JP4298531B2 - Input attribute condition determination device, input attribute condition determination method, input attribute condition determination program, data analysis device, data analysis method, and data analysis program - Google Patents

Description

  The present invention relates to an output attribute (object attribute) to be analyzed, such as characteristics of a product manufactured in a manufacturing process, and an input attribute (description attribute) that is an attribute affecting the output attribute, such as a manufacturing process condition. An input attribute condition determining device, an input attribute condition determining method, an input attribute condition determining program, and an input attribute condition determining program for determining an input attribute condition such that an output attribute value is collected for data composed of The present invention relates to a data analysis apparatus, a data analysis method, and a data analysis program for analyzing a causal relationship between an input attribute and an output attribute using the apparatus.

  A decision tree technique is known as an effective technique for analyzing a causal relationship between an output attribute and an input attribute (see Patent Document 1). This method creates a tree structure in which the values of the output attributes are well organized at the leaf portions that are sequentially separated by the values of the input attributes.

  FIG. 10 is an example of a decision tree described in the prior art section of Patent Document 1 (see paragraphs [0002] to [0005] and FIG. 22 of Patent Document 1). The data group of [Best Mode for Carrying Out the Invention] is a target of analysis. The data group in Table 1 is a set in which twelve pieces of data including a set of four input attribute values x1, x2, x3, and x4 and a value of an output attribute y corresponding to these input attributes are collected. In a decision tree created by this method (hereinafter referred to as “conventional decision tree-1”), as shown in FIG. 10, the values X, Y, and Z of the output attributes y are input attributes x2, x3. , X1 are well separated.

  However, in the creation of the conventional decision tree-1 in FIG. 10, when data is classified, it is classified into data sets corresponding to the number of values that the input attribute takes (the number of attribute value types). For example, since the input attribute x2 takes four types of values (a, b, c, d), it is classified into four sets by classification based on the input attribute x2. Therefore, if the number of values that the input attribute takes increases, the decision tree may become complicated.

  As a solution to this problem, Patent Document 1 proposes a decision tree in which attribute values to be grouped are represented by one label for each attribute, and data is classified by the label.

  FIG. 11 shows a label hierarchy described in an example of Patent Document 1 (see paragraphs [0010] to [0028] and FIG. 13 of Patent Document 1). In this embodiment, for example, for the x3 attribute composed of four types of attribute values (1, 2, 3, 4), the x3 attribute values “1” and “2” are labeled “2.5 or less” and x3 The attribute values “3” and “4” are labeled “2.5 or more” to express the hierarchical structure.

  FIG. 12 is an example of a decision tree created using the label hierarchical structure of FIG. 11 described in Patent Document 1, and the data group of Table 1 is an analysis target (paragraph [0010] to [0010] of Patent Document 1). [0028] and FIG. 14). In the decision tree of FIG. 12, the values X, Y, and Z of the output attribute y are well separated by the values of the input attributes x2, x3, and x1. Also, as shown in FIG. 12, the decision tree created using the label hierarchical structure of FIG. 11 (hereinafter, this decision tree is referred to as conventional decision tree-2) is the conventional decision shown in FIG. Compared to Tree-1, it is very concise.

  Here, when the tree structure as described above is formed, an evaluation index for determining an optimal branching condition is required at each node (t1, t2, t3,... In FIG. 12). As such an evaluation index, the Gini index, the least square criterion, and the like are known, but when the output attribute (target attribute) y is a qualitative variable as in the data of Table 1, the Gini index is Often used. Hereinafter, the Gini index method, which is a method for determining an optimal branch condition using the Gini index, will be described.

  As described in p44 to p47 of Non-Patent Document 1, the Gini index is expressed by the following equation.

i (t) = 1−Σ {p (j | t)} ² (1)
Here, p (j | t) is a probability that the output attribute y is y = j at the node t.

  The small Gini index i (t) means that the values of the output attributes y are well organized at the node t.

As an example, the Gini index at the root node t1 in FIG.
i (t1) = 1−Σ {p (j | t1)} ²
= 1-{(X | t1) ² + (Y | t1) ² + (Z | t1) ² }
= 1-{(4/12) ² + (4/12) ² + (4/12) ² }
= 0.667 (2)
It becomes.

In addition, the Gini index at the child node t2 of the root node t1 where x2 = a or b is
i (t2) = 1−Σ {p (j | t2)} ²
= 1-{(X | t2) ² + (Y | t2) ² + (Z | t2) ² }
= 1-{(4/10) ² + (2/10) ² + (4/10) ² }
= 0.64 (3)
In addition, the Gini index in the child node t3 of the root node t1 where x2 = c or d is
i (t3) = 1−Σ {p (j | t3)} ²
= 1-{(X | t3) ² + (Y | t3) ² + (Z | t3) ² }
= 1-{(0/2) ² + (2/2) ² + (0/2) ² }
= 0 (4)
It becomes. Here, i (t3) = 0 indicates that the values of the output attribute y are well organized under the condition x2 = c or d (the value of the output attribute y is Y only). )

  It can be evaluated based on the above Gini index how much the output attribute is improved by branching the root node t1 to the child nodes t2 and t3. In the Gini index method, an improvement degree Δi (t1) represented by the following formula is used as this evaluation index.

Δi (t1)
= I (t1)-{pt2 · i (t2) + pt3 · i (t3)} (5)
Here, pt2 and pt3 are used to branch the root node t1 (12 data) into a child node t2 (x2 = a or b; 10 data) and a child node t3 (x2 = c or d; 2 data). Represents the branching ratio, and pt2 = 10/12 and pt3 = 2/12.

Therefore, in the example of FIG. 12, the extent to which the set of output attributes is improved by branching the root node t1 into the child nodes t2 and t3 is as follows:
Δi (t1)
= I (t1)-{pt2 · i (t2) + pt3 · i (t3)}
= 0.667-{(10/12) x 0.64 + (2/12) x 0}
= 0.134 (6)
It becomes.

  In Patent Document 1, the degree to which the group of output attributes is improved is evaluated using the following formula (7). However, the basic idea is that the improvement degree of the Gini index method ((5) (6 ) Is the same as formula).

Δi ′ (t1) = Σ {p (j | t2)} ² + Σ {p (j | t3)} ² (7)
The Gini index (Equations (3) and (4)) and the improvement (Equation (6)) are calculated for all patterns of branching conditions that can be taken by each input attribute. Of these, the condition that maximizes the degree of improvement is determined as the final branch condition. In the example of branching from the root node t1 in FIG. 12, the branching conditions of t2: “x2 = a, b” and t3: “x2 = c, d” have the maximum improvement degree Δi (t1). Finally selected.

Here, the number of patterns of the branch condition when calculating the Gini index and the degree of improvement is determined by the number of values that the input attribute can take. For example, for the input attribute x2, the possible values are a and b. , C and d,
T2: “x2 = a”, t3: “x2 = b, c, d”
T2: “x2 = a, b”, t3: “x2 = c, d”
T2: “x2 = a, b, c”, t3: “x2 = d”
It becomes three patterns. The number of patterns increases as the number of values that the input attribute can take increases.

In Patent Document 1, the calculation for determining the branch condition (equation (7)) is performed only for the classification pattern based on the label in FIG. The calculation is simplified.
JP-A-8-314725 (Publication date: November 29, 1996) Atsushi Otaki, Yuji Horie, Dan Steinberg, "Applied binary tree analysis method-by CART", Nikka Giren, July 6, 1998, p44-p47

  Using the conventional Gini index method (Gini index and degree of improvement) for determining the optimum branching condition in the past for the application of factor analysis of product characteristic defects in the manufacturing process of devices such as devices, Explain the problem.

  Now, the input attributes x1, x2, x3, and x4 in Table 1 are various process data and in-line inspection data in the product manufacturing process, and the output attribute y is the product characteristic data. The output attribute y = Y is the product characteristic. It shall correspond to a defect. Then, the process engineer uses the Gini index method for the product characteristic defect y = Y to determine the cause of the product characteristic defect (“Which input attribute is in which value range? Which product characteristic is bad?”). Shall be investigated. In this way, when investigating the causes of product characteristic defects, rather than forming a strict decision tree of a deep hierarchy, an optimal branching condition for separating good products and defective products for each input attribute ( In many cases, it is required to clarify a threshold value) and to extract a condition having a high degree of influence on a defect among the optimum branch conditions of each of these input attributes.

  FIG. 20 shows the improvement degree of the Gini index method calculated for the branch condition pattern that each input attribute can take in order to separate a good product and a defective product (branch from a root node to a child node) using the above-mentioned material. To FIG. The number of branch condition patterns for calculating the Gini index and the improvement degree is a total of 12 conditions (3 conditions for each input attribute) in all the branch condition patterns that the input attributes x1, x2, x3, and x4 can take. From FIG. 20 to FIG. 23, as a branching condition for separating a non-defective product with y = X, Z and a defective product with y = Y, for the input attribute x1, “x1 = A, B and x1 = C, D For the input attribute x2, “branch” is “branch between x2 = a, b and x2 = c, d”, and for the input attribute x3, “x3 = 1, 2 and x3 = 3, 4” It can be seen that “branch” is appropriate for the input attribute x4, “branch between x4 = 10 and x4 = 20, 30, 40”. As described above, when the data group in Table 1 is an analysis target, it is possible to extract a condition (optimal branch condition for each input attribute) that causes a product characteristic defect by the Gini index method.

  However, in the manufacturing site of a product such as an actual device (especially a semiconductor device), there are process data and in-line inspection data of about 10 to 100 attributes per process, and the value is a multi-value with many effective digits. It is a numerical value. For example, the number of values that one input attribute can take is on the order of tens of thousands to hundreds of thousands. In such a case, the same Gini index as in the equations (3) and (4) and the improvement degree similar to that in the equation (6) are calculated tens of thousands to hundreds of thousands of times per attribute. Tens of thousands to hundreds of thousands of calculations are performed for the number of input attributes. Such a large-scale calculation takes an enormous amount of time, and in some cases, the computer's memory is insufficient, which may make the calculation impossible. That is, the conventional Gini index method has a problem that the calculation load is large. Therefore, the determination of the optimal branch condition by the Gini index method and the data analysis using this are inefficient.

Further, as another problem of the Gini index method, when the probability p (Y | t) of a defective product is extremely smaller than the probability p (XZ | t) of a good product,
i (t) = 1−Σ {p (j | t)} ²
= 1- {p (XZ | t) ² + p (Y | t) ² } (8)
In the Gini index represented by, the probability p (Y | t) of the defective product is hardly reflected, and the accuracy in extracting the condition (the optimal branch condition in each input attribute) for separating the non-defective product and the defective product is There was a problem of lowering. This problem becomes prominent when almost no defective products are included in the root node (total sample). However, if attention is paid to a specific defect category, there are not a few cases. Therefore, the accuracy of the determination of the optimal branch condition by the Gini index method and the data analysis using this is low.

  Another problem of the prior art will be described using the case where the decision tree generation method of Patent Document 1 is applied to cause analysis of product characteristic defects in a manufacturing process of a product such as a device.

  Now, the input attributes x1, x2, x3, and x4 in Table 1 are various process data and in-line inspection data in the product manufacturing process, and the output attribute y is the product characteristic data. The output attribute y = Y is the product characteristic. It shall correspond to a defect. Then, the process engineer uses the decision tree-1 (FIG. 10) generated by the technique described in the prior art of Patent Document 1 or the technique described in Patent Document 1 for the product characteristic defect y = Y. It is assumed that the cause of product characteristic failure is investigated using the generated conventional decision tree-2 (FIG. 12).

  At this time, in the decision tree-1 generated by the method described in the prior art of Patent Document 1, y = Y to be noticed is distributed at a plurality of locations (four locations in the example of FIG. 10) in the tree shape. Therefore, it is complicated, and it is difficult for a process engineer to determine the cause of a product characteristic failure such as “Which input attribute is in which value range, so that the product characteristic is bad?”. In the example of FIG. 10, since there are only four input attributes and only four types of attribute values, the process engineer can somehow determine the cause of the product characteristic failure. However, in the manufacturing site of a product such as an actual device (especially a semiconductor device), there are process data and in-line inspection data of about 10 to 100 attributes per process, and the value is a multivalue with a large number of significant digits. It is distributed in a very wide range. For example, the number of values that one input attribute can take is on the order of tens of thousands to hundreds of thousands. Furthermore, due to the influence of disturbance (attributes that cannot be detected as input attributes), the values of output attributes often vary even if the values of the input attributes are the same. In such a case, if the method described in the prior art of Patent Document 1 is used, a precise analysis is aimed at, but the data is classified into an infinite number of data sets. It becomes impossible to specify the cause of the characteristic failure.

  On the other hand, in the decision tree-2 (FIG. 12) generated by the method disclosed in Patent Document 1, the decision tree is simple because the classification is based on the label hierarchy. Therefore, it is easy for the process engineer to identify the cause of the product characteristic failure where y = Y.

  However, in order to create the simple decision tree-2 shown in FIG. 12, it is necessary to previously define the label hierarchical structure shown in FIG. For this reason, the decision tree generation method of Patent Document 1 cannot be applied when there is no idea of the attribute values to be collected. As described above, in the manufacturing site of a product such as an actual device, there are process data and in-line inspection data having about 10 to 100 attributes per process, and the value is a multi-value having a large number of significant digits. It is numerical and distributed in a very wide range. Furthermore, due to the influence of disturbance (attributes that cannot be detected as input attributes), the values of output attributes often vary even if the values of the input attributes are the same. Under these circumstances, it is very difficult even for an experienced process engineer to find an attribute value that is collected as one label for each input attribute. Therefore, the data analysis of Patent Document 1 is inefficient.

  The present invention has been made in view of the above-described conventional problems, and an object thereof is to improve the efficiency of data analysis. More specifically, an object of the present invention is to provide an input attribute condition determination apparatus and an input attribute that can improve data analysis efficiency by greatly reducing the calculation load when determining the condition of the input attribute that separates a good product and a defective product. Condition determination method and input attribute condition determination program, data analysis apparatus and data capable of improving data analysis efficiency by deriving causal relationship between output attribute and input attribute in a concise form without predefining label hierarchy To provide an analysis method and a data analysis program.

  In view of the above problems, another object of the present invention is to provide an input attribute condition determination device, an input attribute condition determination method, and an input attribute condition determination capable of determining with high accuracy an input attribute condition for separating a non-defective product from a defective product. To provide a program.

  In order to solve the above problem, an input attribute condition determination device according to the present invention is a set of data composed of at least one input attribute that is a numerical attribute and an output attribute, and depends on the value of the output attribute. Conditions for input attributes for dividing the analysis data group into two so that the first data group and the second data group are grouped with respect to the analysis data group classified into the first data group and the second data group An input attribute condition determining apparatus for determining an input attribute condition, wherein, for each of all numerical values taken by the input attribute, a ratio of data having an input attribute equal to or lower than the numerical value in the first data group is first set. Frequency calculation means for calculating, as the second frequency, the proportion of data whose input attribute is less than or equal to the numerical value in the second data group, and each of all the numerical values taken by the input attribute. The difference calculating means for calculating the difference value between the first frequency and the second frequency, and among the numerical values taken by one input attribute, the numerical value that maximizes the difference value is the threshold value in the input attribute. And threshold value determining means for determining at least one threshold value corresponding to at least one input attribute, and input attribute condition determining means for determining the input attribute condition based on the threshold value determined by the threshold value determining means. It is characterized by that.

  In order to solve the above-described problem, an input attribute condition determination method according to the present invention uses the input attribute condition determination device, and includes data composed of at least one input attribute that is a numerical attribute and an output attribute. The analysis data is such that the first data group and the second data group are grouped with respect to the analysis data group classified into the first data group and the second data group according to the value of the output attribute. An input attribute condition determination method for determining an input attribute condition, which is an input attribute condition for bisecting a group, wherein the first data group is obtained for each of all numerical values taken by the input attribute by the frequency calculation means. The ratio of data whose input attribute is less than or equal to the numerical value is calculated as the first frequency, and the ratio of data whose input attribute is equal to or less than the value in the second data group is the second frequency. A frequency calculating step for calculating, a difference calculating step for calculating a difference value between the first frequency and the second frequency for each of all numerical values taken by the input attribute by the difference calculating means, and the threshold determining means A threshold value determining step of determining a numerical value that maximizes the difference value among the numerical values that one input attribute takes as a threshold value in the input attribute, and determining at least one threshold value corresponding to at least one input attribute; And an input attribute condition determining step for determining the input attribute condition based on the threshold value determined by the threshold value determining means by the input attribute condition determining means.

  An input attribute condition determination program according to the present invention is a set of data composed of at least one input attribute that is a numerical attribute and an output attribute in order to solve the above-described problem, and depends on the value of the output attribute. With respect to the analysis data group classified into the first data group and the second data group, the computer has the input attribute for each of all the numerical values that the input attribute takes in the first data group that is less than or equal to the numerical value. The frequency calculation means for calculating the data ratio as the first frequency and calculating the ratio of the data whose input attribute is equal to or lower than the numerical value in the second data group as the second frequency, all numerical values taken by the input attribute Difference calculation means for calculating a difference value between the first frequency and the second frequency, and among the numerical values taken by one input attribute, the numerical value that maximizes the difference value is used as the input attribute. Oh Threshold value determining means for determining at least one threshold value corresponding to at least one input attribute, and the first data group and the second data group based on the threshold value determined by the threshold value determining means respectively. It is an input attribute condition determining program for functioning as an input attribute condition determining means for determining an input attribute condition, which is an input attribute condition for dividing the analysis data group into two groups.

  In order to solve the above problems, a computer-readable recording medium according to the present invention records the above-mentioned input attribute condition determination program.

  According to the above device, method, program, or recording medium, a difference value between the first frequency and the second frequency is calculated for each of all numerical values taken by the input attribute, and this difference value is calculated as the first data. The analysis data group is used as a threshold evaluation index to divide the analysis data group into two so that the group and the second data group are combined. Then, among the numerical values taken by one input attribute, the numerical value that maximizes the evaluation index (difference value) is set as the threshold value of the input attribute, and at least one threshold value corresponding to at least one input attribute is determined. Yes. As a result, the threshold evaluation corresponding to the improvement degree of the Gini index method can be achieved by a very simple calculation process only for calculating the difference value between the first frequency and the second frequency for each of the numerical values taken by the input attribute. An indicator can be obtained. That is, in the above configuration, like the Gini index method, the Gini index (Equation (3) (4)) and the improvement level (Equation (6)) are calculated for every pattern of all branch conditions that can be taken by the input attribute. It is not necessary to perform such an enormous amount of arithmetic processing, and it is only necessary to perform arithmetic processing for obtaining a difference value for data corresponding to the number of values that the input attribute can take. Therefore, the number of values that one input attribute can take is on the order of tens of thousands to hundreds of thousands, as in the case where the analysis data group is data of a manufacturing process of a product such as an actual device (especially a semiconductor device). Even in such a case, almost no calculation load is applied, and processing can be performed in a short time. That is, it is possible to determine an input attribute condition (optimal branch condition for each input attribute) that separates the first data group and the second data group in a short time without requiring a calculation load. Therefore, the efficiency of data analysis can be improved.

  Further, according to the above configuration, the first frequency and the second frequency are respectively normalized by the total number of data in the data group in the corresponding data group, the number of data whose input attribute is equal to or less than the numerical value. Therefore, even if the ratio of the first data group and the ratio of the second data group in the analysis data group are extremely different, the difference value does not decrease the accuracy. By using the value as a threshold evaluation index, it is possible to determine the input attribute condition (the optimal branch condition for each input attribute) for separating the first data group and the second data group with high accuracy.

  The input attribute condition determining apparatus according to the present invention further includes polarity determining means for determining a magnitude relationship between the first frequency and the second frequency in the threshold determined by the threshold determining means, and the input attribute condition determining The means determines the first frequency as the second frequency by the polarity determining means so that the second data group is grouped into the data group that satisfies the input attribute condition and the first data group is grouped into the data group that does not satisfy the input attribute condition. If it is determined that the input attribute condition is greater than the first attribute, the input attribute condition is determined to be a condition that the input attribute exceeds the threshold, and the polarity determination unit determines that the second frequency is greater than the first frequency. More preferably, the input attribute condition is determined as a condition that “input attribute is equal to or less than threshold”. As a result, an input attribute condition in which the second data group is grouped into a data group that satisfies the input attribute condition, and the first data group is grouped into a data group that does not satisfy the input attribute condition, It can be determined to be a condition “exceeding” or a condition “input attribute is equal to or less than a threshold”.

  In one embodiment of the input attribute condition determining apparatus according to the present invention, the input attribute is a manufacturing process condition and / or inline inspection result in a product manufacturing process, and the output attribute is a product quality determination result. In addition, the second data group is a data group with a poor quality determination result.

  In this case, it is possible to specify the manufacturing process conditions in the manufacturing process and the characteristics (in-line inspection result) during the manufacturing, which are factors that cause a defective product (a product whose quality judgment result is defective).

  In order to solve the above problem, the data analysis apparatus according to the present invention provides an input attribute and an output attribute for a basic data group that is a set of data composed of a plurality of input attributes and output attributes. A data analysis apparatus for analyzing a causal relationship and extracting information indicating the causal relationship, classifying the basic data group into a first data group and a second data group based on an output attribute value, and a classification flag A classifying means for assigning, an analysis data group extracting means for extracting an analysis data group to be analyzed from the basic data group after the classification, and an input attribute condition determining apparatus according to claim 1 or 2 The frequency calculation means and the difference calculation means perform the calculation for each of all the numerical values taken by each input attribute of the analysis data group, and the threshold value determination means for each input attribute of the analysis data group, That It is characterized by determining the threshold value.

  In order to solve the above problems, a data analysis method according to the present invention uses the data analysis apparatus described above to generate a basic data group that is a set of data composed of a plurality of input attributes and output attributes. A data analysis method for analyzing a causal relationship between an input attribute and an output attribute and extracting information indicating the causal relationship, wherein the basic data group is first determined by the classification means according to the value of the output attribute. The analysis data group to be analyzed is extracted from the basic data group after the classification by the classification step for classifying the data group and the second data group and assigning the classification flag, and the analysis data group extraction means. In the first data group of the analysis data group, for each of all the numerical values taken by each input attribute of the analysis data group by the analysis data group extraction step to be performed and the frequency calculation means of the input attribute condition determination device The ratio of data whose input attribute is less than or equal to the numerical value is calculated as the first frequency, and the ratio of data whose input attribute is equal to or less than the value in the second data group of the analysis data group is calculated as the second frequency. The difference between the first frequency and the second frequency for each of all the numerical values taken by each input attribute of the analysis data group by the frequency calculation step calculated as follows and the difference calculation means of the input attribute condition determination device A difference calculating step for calculating a value, and a threshold value determining step for determining, for each input attribute, a numerical value that maximizes the difference value for each input attribute by the threshold value determining means of the input attribute condition determining device. And the first data group and the second data based on the threshold value determined by the threshold value determining means by the input attribute condition determining means of the input attribute condition determining device. Bets is characterized in that it comprises an input attribute condition determining step of determining an input attribute conditions for 2 differentiating the analytical data group as settled respectively.

  In order to solve the above problems, a data analysis program according to the present invention provides a computer to a basic data group that is a set of data composed of a plurality of input attributes and output attributes. Are classified into the first data group and the second data group according to the value of the output attribute, the classification means for assigning the classification flag, and the analysis data to be analyzed from the basic data group after the classification Analytical data group extraction means for extracting a group, and for each of all numerical values taken by each input attribute of the analytical data group, in the first data group of the analytical data group, the ratio of data whose input attribute is less than or equal to the numerical value A frequency calculation means for calculating the first frequency and calculating a ratio of data whose input attribute is equal to or lower than the numerical value in the second data group of the analysis data group as the second frequency, and the analysis data group Difference calculation means for calculating a difference value between the first frequency and the second frequency for each of all numerical values taken by each input attribute, and for each input attribute, a numerical value that maximizes the difference value. A threshold value determining means for determining the threshold value of the input attribute, and for dividing the analysis data group into two so that the first data group and the second data group are grouped based on the threshold value determined by the threshold value determining means. It is a data analysis program for functioning as an input attribute condition determining means for determining an input attribute condition that is a condition of the input attribute.

  In order to solve the above problems, a computer-readable recording medium according to the present invention records the above data analysis program.

  According to the apparatus, method, program, or recording medium, since the input attribute condition determining apparatus, method, program, or recording medium is included, the efficiency of data analysis can be improved, and the first data group An input attribute condition (optimal branch condition for each input attribute) that separates the second data group can be determined with high accuracy.

  The data analysis apparatus according to the present invention may further include a numerical value conversion means for performing processing for converting an input attribute into a numerical value for a basic data group including an input attribute that is not a numerical attribute. This makes it possible to determine the input attribute condition for a basic data group that is a set of data composed of at least one non-numeric input attribute and output attribute.

  In the data analysis apparatus according to the present invention, the input attribute condition determination device is configured to determine a plurality of input attribute conditions. For each of the input attribute conditions determined by the input attribute condition determination device, an “input” A division rule evaluation means for calculating a division rule evaluation value representing the probability of the association rule that if the attribute satisfies the input attribute condition, the data is included in the second data group in the analysis data group; and the input attribute condition Based on the input attribute condition having the largest division rule evaluation value among the input attribute conditions determined by the determination device, the analysis data group is divided into the factor data group satisfying the input attribute condition and the input attribute condition. A dividing unit that divides the data into other data groups that are not satisfied may be included.

  In the data analysis method according to the present invention, for each of the input attribute conditions determined by the input attribute condition determination device by the division rule evaluation unit, “if the input attribute satisfies the input attribute condition, A division rule evaluation step for calculating a division rule evaluation value representing the probability of an association rule that is "data included in two data groups", and an input attribute condition determined by the input attribute condition determination device by the division means And dividing the analysis data group into a factor data group satisfying the input attribute condition and another data group not satisfying the input attribute condition based on the input attribute condition having the largest division rule evaluation value Steps may be included.

  According to the above apparatus and method, among the plurality of input attribute conditions determined by the input attribute condition determining apparatus, the factor data group satisfying the input attribute condition having the maximum division rule evaluation value, that is, the second data group A data group having the largest factor (input attribute condition) that causes a corresponding problem event (for example, occurrence of defective product) can be derived.

  In the data analysis apparatus according to the present invention, the analysis data group extraction unit extracts at least one of the data groups divided by the division unit as a new analysis data group, and performs processing and input by the analysis data group extraction unit. It is preferable that a series of processes including a process by the attribute condition determination device, a process by the division rule evaluation unit, and a process by the division unit are repeatedly executed.

  According to the above configuration, more detailed factor analysis results can be obtained through repeated processing, and a tree structure can be created with a plurality of factors as nodes. Therefore, even if the analysis target is intertwined with a plurality of factors that are difficult to express with a single association rule, the factors can be determined with sufficiently high accuracy.

  Further, when the repeated process is not performed, even if the accuracy of the threshold evaluation index (difference value) is low due to the influence of disturbance, this problem can be solved by performing the repeated process.

  Furthermore, in a certain input attribute, the output attribute condition factor corresponding to the second data group is repeated even when the two types are “input attribute is below threshold” and “input attribute exceeds threshold”. Both of these factors can be extracted by this process.

  In the data analysis apparatus according to the present invention, it is preferable that the analysis data group extraction unit extracts only another data group from the data group divided by the division unit as a new analysis data group.

  According to the above configuration, since only the other data group among the data groups divided by the dividing unit is subjected to the above-described repetitive processing as a new analysis data group, the cause of the output attribute condition corresponding to the second data group A simple and sufficient factor analysis result is obtained for the analysis.

  In addition, since the other data group is processed as a new analysis data group, the influence of the factor (input attribute condition) extracted in the process of the previous repeated processing can be excluded, and the output corresponding to the second data group New factors of attribute conditions can be extracted with high accuracy.

  The division rule evaluation means, for each of the input attribute conditions determined by the input attribute condition determination device, for the ratio of data satisfying the input attribute condition in the first data group of the analysis data group It is preferable that the ratio of the ratio of data satisfying the input attribute condition in the second data group is calculated as a division rule evaluation value. Thereby, a rule evaluation value can be easily calculated.

  The data analysis apparatus according to the present invention further includes a classification condition setting means for setting a classification condition, and the classification means classifies the basic data group based on the classification condition set by the classification condition setting means. May be. As a result, the user can arbitrarily set the classification condition, so that various input attribute conditions (factors) corresponding to the classification condition can be derived.

  In the data analysis apparatus according to the present invention, the basic data group includes a plurality of output attributes, the classification condition setting unit sets a classification condition for each of the plurality of output attributes, and the classification unit includes: The basic data group may be classified according to the logical sum or logical product of the respective classification conditions set by the classification condition setting means. Accordingly, it is possible to derive a factor that satisfies a plurality of output attribute conditions or a factor that satisfies any of the plurality of output attribute conditions.

  In the data analysis device according to the present invention, the input attribute condition determination device is configured to determine a plurality of input attribute conditions. For each of the input attribute conditions determined by the input attribute condition determination device, the basic attribute A second data group separation degree calculating means for calculating a second data group separation degree representing a ratio of the second data group included in the data satisfying the input attribute condition in the data group; and the input attribute condition determining device. Among the input attribute conditions determined in (2), an input attribute condition having a second data group separation degree having a value larger than the second data group content ratio representing the ratio of the second data group in the basic data group, A configuration including first factor extracting means for extracting as information indicating a factor of the output attribute condition corresponding to the second data group may be employed.

  According to the above configuration, the second data group separation is performed not only for the input attribute condition (branch condition in the decision tree) that is the largest factor of the output attribute condition corresponding to the second data group, but also for other input attribute conditions. All high-level input attribute conditions can be extracted. Therefore, even if there is a competing factor (competitive factor) in the maximum factor (branch condition in the decision tree) of the output attribute condition corresponding to the second data group, it can be reliably captured without missing the factor. Further, according to the above configuration, for the plurality of determined factors (input attribute conditions), the second data group separation degree is used as an evaluation index, and the priority order (the order of the influence degree on the output attribute condition corresponding to the second data group) ) Can be attached.

  Furthermore, according to the above configuration, the information indicating the factor of the output attribute condition is extracted based on the clear index of the second data group separation degree indicating the probability that the second data group is separated from the first data group. Yes. Therefore, it is possible to clearly grasp the cause of the problem event regardless of how complicated the decision tree is.

  In the data analysis device according to the present invention, the input attribute condition determining device is configured to determine a plurality of input attribute conditions, and the largest of the input attribute conditions determined by the input attribute condition determining device. A configuration including a second factor extracting unit that extracts an input attribute condition having a division rule evaluation value as information indicating a factor of an output attribute condition corresponding to the second data group may be employed.

  According to the above configuration, the factor of the output attribute condition (result) corresponding to the second data group can be extracted in a simple manner without defining the label hierarchical structure in advance, and the efficiency of data analysis can be improved. .

  In order to solve the above problem, the data analysis apparatus according to the present invention provides an input attribute and an output attribute for a basic data group that is a set of data composed of a plurality of input attributes and output attributes. A data analysis apparatus for analyzing a causal relationship and extracting information indicating the causal relationship, classifying the basic data group into a first data group and a second data group based on an output attribute value, and a classification flag All the input that can be taken by each input attribute of the analysis data group and the analysis data group extraction means for extracting the analysis data group to be analyzed from the basic data group after the classification For each attribute condition, “if the input attribute satisfies the input attribute condition, the data belongs to the second data group in the analysis data group, and if the input attribute does not satisfy the input attribute condition, the first in the analysis data group. Belonging to data group The first evaluation means for calculating the input attribute condition evaluation index, which represents the certainty of the first association rule that is “the data to be received”, and the maximum input attribute condition evaluation for each input attribute of the analysis data group An input attribute condition determining unit that determines an input attribute condition having an index as an input attribute condition satisfying the first correlation rule, and for each of the input attribute conditions determined by the input attribute condition determining unit, A second evaluation means for calculating a second evaluation index representing the probability of the second association rule that the data is included in the second data group in the analysis data group if the input attribute condition is satisfied; Among the input attribute conditions determined by the attribute condition determining means, the input attribute condition that maximizes the second evaluation index is used as information indicating the cause of the output attribute condition corresponding to the second data group. It is characterized in that it comprises a second factor extraction means for output.

  In order to solve the above problems, a data analysis method according to the present invention uses the data analysis apparatus described above to generate a basic data group that is a set of data composed of a plurality of input attributes and output attributes. A data analysis method for analyzing a causal relationship between an input attribute and an output attribute and extracting information indicating the causal relationship, wherein the basic data group is first determined by the classification means according to the value of the output attribute. The analysis data group to be analyzed is extracted from the basic data group after the classification by the classification step for classifying the data group and the second data group and assigning the classification flag, and the analysis data group extraction means. For each of all the input attribute conditions that can be taken by each input attribute by the analysis data group extracting step and the first evaluation means, “if the input attribute satisfies the input attribute condition, the second in the analysis data group Data attribute belonging to the data group, and if the input attribute does not satisfy the input attribute condition, it is data belonging to the first data group in the analysis data group. By the first evaluation step for calculating the evaluation index and the input attribute condition determining means, the input attribute condition having the maximum input attribute condition evaluation index for each input attribute satisfies the first correlation rule. For each of the input attribute conditions determined by the input attribute condition determining means by the input attribute condition determining step determined as the input attribute condition and the second evaluation means, “if the input attribute satisfies the input attribute conditions, the analysis The second evaluation step for calculating the second evaluation index, which represents the probability of the second association rule that the data is included in the second data group in the data group. And the input attribute condition that maximizes the second evaluation index among the input attribute conditions determined by the input attribute condition determining means by the second factor extracting means is the output attribute corresponding to the second data group. And a second factor extracting step for extracting as information indicating the factor of the condition.

  In order to solve the above problems, a data analysis program according to the present invention provides a computer to a basic data group that is a set of data composed of a plurality of input attributes and output attributes. Are classified into the first data group and the second data group according to the value of the output attribute, and the classification means for assigning the classification flag, the analysis data group to be analyzed from the basic data group after the classification Analytical data group extracting means for extracting, for each of all the input attribute conditions that each input attribute can take, “if the input attribute satisfies the input attribute condition, the data belongs to the second data group in the analytical data group If the input attribute does not satisfy the input attribute condition, it is data belonging to the first data group in the analysis data group. The first evaluation means, for each input attribute, the input attribute condition determining means for determining the input attribute condition having the maximum input attribute condition evaluation index as the input attribute condition satisfying the first correlation rule, and the input For each of the input attribute conditions determined by the attribute condition determining means, the confirmation of the second correlation rule that “if the input attribute satisfies the input attribute condition, it is data included in the second data group in the analysis data group”. A second evaluation means for calculating a second evaluation index representing the likelihood, and the input attribute condition that maximizes the second evaluation index among the input attribute conditions determined by the input attribute condition determination means is the second data It is a data analysis program for functioning as a second factor extracting means for extracting as information indicating the factor of the output attribute condition corresponding to the group.

  In order to solve the above problems, a computer-readable recording medium according to the present invention records the above data analysis program.

  According to the apparatus, method, program, or recording medium, the factor of the output attribute condition (result) corresponding to the second data group can be extracted in a concise form without defining the label hierarchical structure in advance. Therefore, for example, if the second data group is a data group corresponding to a bad result (for example, occurrence of a defective product), the user can easily grasp the cause of the bad result. Conversely, if the second data group is a data group corresponding to a good result (for example, occurrence of a product having excellent characteristics), the user can easily grasp the factor of the good result. Therefore, the efficiency of data analysis can be improved.

  In the data analysis device, the first evaluation unit is configured to perform, for each of all the numerical values of each input attribute, by bifurcation of data in which the input attribute is equal to or lower than the numerical value and data in which the input attribute exceeds the numerical value. A threshold evaluation index representing the degree of separation between one data group and the second data group is calculated as the input attribute condition evaluation index, and the second evaluation means is the input attribute determined by the input attribute condition determination means For each condition, the ratio of the ratio of data satisfying the input attribute condition in the second data group of the analysis data group to the ratio of data satisfying the input attribute condition in the first data group of the analysis data group More preferably, the second evaluation index is calculated.

  In order to solve the above problems, a data analysis apparatus according to the present invention analyzes a basic data group, which is a set of data composed of a plurality of input attributes and output attributes, and includes input attributes and output attributes. Is a data analysis apparatus for analyzing the cause-and-effect relationship and extracting information indicating the cause-and-effect relationship, and classifying means for classifying the basic data group into a first data group and a second data group according to output attributes, and each input A first evaluation means for calculating a threshold evaluation index representing a degree that data having an input attribute equal to or less than the numerical value is biased to one of the first data group and the second data group for all the numerical values of the attribute; Based on the threshold evaluation index calculated by the first evaluation means, threshold determination means for determining a numerical value having the maximum threshold evaluation index for each input attribute as a threshold of each input attribute, and each determined by the threshold determination means Input genus The first rule evaluation value indicating the probability of the correlation rule “if the input attribute is equal to or less than the threshold, the data is included in the second data group”, and “the input attribute exceeds the threshold The second rule evaluation value representing the likelihood of the association rule that the data is included in the second data group for each input attribute, and the rule evaluation by the second evaluation means The data indicating the input attribute condition of the correlation rule having the highest rule evaluation value among the correlation rules related to all the input attributes whose values have been calculated is used as information indicating the cause of the output attribute condition corresponding to the second data group And a second factor extracting means for extracting.

  According to the above configuration, the factor of the output attribute condition (result) corresponding to the second data group can be extracted in a simple manner without defining the label hierarchical structure in advance, and the efficiency of data analysis can be improved. .

  The data analysis apparatus according to the present invention includes an analysis data group based on the input attribute condition extracted by the second factor extraction means, a factor data group that satisfies the input attribute condition, and does not satisfy the input attribute condition. The analysis data group extraction means further includes at least one of the data groups divided by the division means as a new analysis data group, and the analysis data group extraction means A series of processing consisting of processing by the first evaluation means, processing by the input attribute condition determination means, processing by the second evaluation means, processing by the second factor extraction means, and processing by the dividing means is repeatedly executed. It is more preferable that it is adapted.

  According to the above configuration, more detailed factor analysis results can be obtained through repeated processing, and a tree structure can be created with a plurality of factors as nodes. Therefore, even if the analysis target is intertwined with a plurality of factors that are difficult to express with a single association rule, the factors can be determined with sufficiently high accuracy.

  In the data analysis apparatus according to the present invention, it is preferable that the analysis data group extraction unit extracts only another data group from the data group divided by the division unit as a new analysis data group.

  According to the above configuration, since only the other data group among the data groups divided by the dividing unit is subjected to the above-described repetitive processing as a new analysis data group, the cause of the output attribute condition corresponding to the second data group A simple and sufficient factor analysis result is obtained for the analysis.

  In addition, since the other data group is processed as a new analysis data group, the influence of the factor (input attribute condition) extracted in the process of the previous repeated processing can be excluded, and the output corresponding to the second data group New factors of attribute conditions can be extracted with high accuracy.

  The first evaluation means calculates, as the first frequency, the ratio of data in which the input attribute is less than or equal to the numerical value in the first data group for all numerical values of each input attribute, and in the second data group , The frequency calculation means for calculating the ratio of data whose input attribute is less than or equal to the numerical value as the second frequency, and the difference between the first frequency and the second frequency for all the numerical values of each input attribute More preferably, difference calculating means is included. Thereby, the threshold evaluation index can be easily calculated. That is, it is possible to determine an input attribute condition (optimal branch condition for each input attribute) that separates the first data group and the second data group in a short time without requiring a calculation load. Further, even if the ratio of the first data group and the ratio of the second data group in the analysis data group are extremely different, the accuracy is not reduced, and the first data group and the second data group are The input attribute condition to be carved (the optimum branch condition for each input attribute) can be determined with high accuracy.

  The second evaluation means uses, as the first rule evaluation value, the ratio of data whose input attribute is equal to or less than the threshold in the second data group to the ratio of data whose input attribute is equal to or less than the threshold in the first data group. Is calculated as the first ratio, and as the second rule evaluation value, the data whose input attribute exceeds the threshold in the second data group with respect to the ratio of the data whose input attribute exceeds the threshold in the first data group More preferably, the ratio is calculated as the second ratio. Thereby, the first and second rule evaluation values can be easily calculated.

  The data analysis apparatus according to the present invention further includes an end condition determination unit that determines whether or not an end condition is satisfied, and the end condition determination unit includes a second analysis data group extracted by the analysis data group extraction unit. Whether the number of data in the data group is 0 is determined as an end condition, and when the end condition determination unit determines that the end condition is satisfied, the execution of the series of processes is ended. It is more preferable. Thereby, it is possible to avoid performing unnecessary processing more than necessary.

  According to the input attribute condition determination device, the input attribute condition determination method, the input attribute condition determination program, and the recording medium on which the input attribute condition determination method of the present invention is recorded, the input attribute condition (each of which separates the first data group and the second data group) The difference value between the first frequency and the second frequency is used as an evaluation index for determining the optimal branch condition in the input attribute), and only the difference value between the first frequency and the second frequency is calculated. The evaluation index (difference value) can be calculated by simple calculation processing. Therefore, it is possible to determine an input attribute condition (optimal branch condition for each input attribute) that separates the first data group and the second data group in a short time without requiring a calculation load.

  In addition, since the first frequency and the second frequency are standardized by the total number of data in the data group corresponding to each, the difference value is the ratio of the first data group to the total data. Even if the ratio of the second data group to all the data is extremely different, it is an accurate evaluation index of the input attribute condition (the optimum branch condition in each input attribute). Therefore, the input attribute condition (optimal branch condition for each input attribute) for separating the first data group and the second data group can be determined with high accuracy.

  According to the data analysis apparatus, the data analysis method, the data analysis program and the recording medium on which the data analysis program of the present invention is recorded, the input attribute condition determination apparatus, method, program, or recording medium is included. In addition, the input attribute condition for separating the first data group and the second data group (the optimal branch condition for each input attribute) can be determined with high accuracy, and the reliability of data analysis can be improved. it can.

  In addition, without pre-defining the label hierarchy, the problem event can be expressed in a very simple form such as “input attribute satisfies the input attribute condition”, for example, “input attribute is below threshold” or “input attribute exceeds threshold”. It is possible to derive a factor that causes a specific output attribute condition (problem event).

  From the above, the present invention has an effect that the efficiency of data analysis can be improved. Further, the input attribute determination device, the input attribute determination method, the input attribute determination program, and the recording medium on which the input attribute determination method of the present invention can determine the input attribute conditions for separating good products from defective products with high accuracy. There is an effect.

[Embodiment 1]
Next, an embodiment of the present invention will be described below.

  First, the data analysis apparatus 100 of this embodiment and the input attribute condition determination apparatus 100A that is a component thereof will be described with reference to FIG. As shown in FIG. 13, the data analysis apparatus 100 includes a basic data group storage unit 102, a character-numeric data conversion unit (numeric conversion unit) 1, a classification condition setting unit (classification condition setting unit) 103, and a data classification unit (classification). Means) 104, basic data group storage unit 105 after classification, analysis data group extraction unit (analysis data group extraction unit) 106, data row separation unit 107, data string extraction unit 5, frequency calculation unit (frequency calculation unit) 6, frequency Cumulative difference calculation unit (difference calculation unit) 7, threshold determination unit (threshold determination unit) 130, polarity determination unit (polarity determination unit) 131, input attribute condition determination unit (input attribute condition determination unit) 111, defective product separation degree calculation Part (second data group separation degree calculating means) 112, first factor extracting part (first factor extracting means) 109, frequency cumulative ratio calculating part (division rule evaluating means) 16, data dividing part (dividing means) 1 5, an end condition determination unit (end condition determination unit) 11, a factor determination unit 117, a complex factor defect number calculation unit 118, a numerical value-character data conversion unit 119, an analysis result data storage unit 14, and an output unit 15. .

  In the data analysis apparatus 100 (FIG. 13), the data row separation unit 107, the data string extraction unit 5, the frequency calculation unit (frequency calculation unit) 6, the frequency cumulative difference calculation unit (difference calculation unit) 7, and the threshold value determination The unit (threshold determination unit) 130, the polarity determination unit (polarity determination unit) 131, and the input attribute condition determination unit (input attribute condition determination unit) 111 constitute an input attribute condition determination device 100A.

  The basic data group storage unit 102 has the same function as the analysis target data storage unit 2 in the data analysis apparatus according to the second embodiment, and is a storage device such as a hard disk that stores the basic data group DA.

  As in the second embodiment, the character-numeric data conversion unit 1 performs input attributes x1, x2, which are numerical attributes, on a basic data group DA which is a set of data composed of non-numeric input attributes and output attributes. A process of converting input attributes into numerical values is performed so that a basic data group DA0 that is a set of data composed of x3, x4 and output attribute y is obtained.

The classification condition setting unit 103 replaces the threshold value setting unit 3 in the data analysis apparatus according to the second embodiment, and sets the classification condition of the output attribute y instead of the threshold value y _th of the output attribute y.

  The data classification unit 104, the basic data group storage unit 105 after classification, and the data row separation unit 107 replace the data classification unit 4 in the data analysis apparatus according to the second embodiment.

  Based on the classification condition set by the classification condition setting unit 103, the data classifying unit 104 determines the basic data group DA0 based on the value of the output attribute y and the first data group DA1 for good products and the second data for defective products. The data is classified into the group DA2, and a classification flag indicating the classification result is assigned to each data.

  The post-classification basic data group storage unit 105 is a storage device such as a hard disk that stores the basic data group (DA00) to which the classification flag is assigned.

  The analysis data group extraction unit 106 extracts an analysis data group DA00 'to be analyzed from the classified basic data group DA00. The analysis data group extraction unit 106 extracts another data group from the data group divided by the data division unit 115 as the next analysis data group DA00 '(new analysis data group). The analysis data group extraction unit 106 may extract all the data groups divided by the data division unit 115 as the next analysis data group DA00 '.

  The data row separation unit 107 divides the data of the analysis data group DA00 'into two based on the respective classification flags, and extracts the first data group DA1 for good products and the second data group DA2 for defective products.

  The data string extraction unit 5 extracts a 1-xj data group that is each data string of the input attribute xj from the first non-defective data group DA1, and also inputs the input attribute xj from the second data group DA2 of defective products. The 2-xj data group which is each data string is extracted.

  The frequency calculation unit 6 and the frequency cumulative difference calculation unit 7 have the same functions as the frequency calculation unit 6 and the frequency cumulative difference calculation unit 7 in the data analysis apparatus of the second embodiment.

  The frequency calculation unit 6 uses the 1-xj data group and the 2-xj data group extracted by the data string extraction unit 5, and uses the input attribute xj for each numerical value of the input attribute xj in the first non-defective data group. The first frequency (1-xj frequency cumulative%) that is the ratio of the number of data that is less than or equal to the numerical value and the ratio of the number of data that the input attribute xj is less than or equal to the numerical value in the second data group of defective products A certain second frequency (2-xj frequency cumulative%) is calculated.

  The frequency cumulative difference calculation unit 7 calculates an xj frequency cumulative difference% representing a difference between the 1-xj frequency cumulative% and the 2-xj frequency cumulative% for each value of the input attribute xj.

  The threshold value determination unit 130 has the same function as the input attribute threshold value determination unit 8 in the data analysis apparatus of the second embodiment, and for each input attribute xj, xj frequency accumulation for each value of the input attribute xj, respectively. The input attribute xj for dividing the analysis data group DA00 ′ into two parts so that the first data group DA1 and the second data group DA2 are combined with the value of the input attribute xj having the maximum value in the difference%. The threshold value xj-th is determined.

  The polarity determination unit 131 determines the magnitude relationship between the first frequency (1-xj frequency cumulative%) and the second frequency (2-xj frequency cumulative%) in the threshold value xj-th determined by the threshold value determination unit 130. Judgment.

  The input attribute condition determining unit 111 collects the defective second data group DA2 in the data group that satisfies the input attribute condition in the analysis data group DA00 ′, and sets the nondefective first data group DA1 in the data group that does not satisfy the input attribute condition. When the polarity determination unit 131 determines that the first frequency is greater than the second frequency, the input attribute condition is determined as a condition that “the input attribute xj exceeds the threshold value xj−th”. When the polarity determination unit 131 determines that the second frequency is greater than the first frequency, the input attribute condition is determined to be a condition that “the input attribute xj is equal to or less than the threshold value xj−th”.

  For each of the input attribute conditions determined by the input attribute condition determination unit 111, the defective product separation degree calculation unit 112 includes second data of defective products in the data satisfying the input attribute condition in the classified basic data group DA00. The defective product separation degree representing the proportion of the group DA2 is calculated.

  The first factor extraction unit 109 is larger than the defective content rate indicating the ratio of the second data group DA2 in the post-classification basic data group DA00 among the input attribute conditions determined by the input attribute condition determination unit 111. The input attribute condition having the value defective product separation degree is extracted as information indicating the factor of the output attribute condition corresponding to the second data group DA2 of the defective product.

  Similar to the second embodiment, the frequency cumulative ratio calculation unit 16 calculates a rule evaluation value for each of the input attribute conditions determined by the input attribute condition determination unit 111. However, the frequency cumulative ratio calculation unit 16 of the present embodiment does not calculate two types of rule evaluation values as in the second embodiment, but instead calculates a frequency cumulative lower ratio or a frequency cumulative upper ratio (described later) as “input. If the attribute satisfies the input attribute condition, it is calculated as a division rule evaluation value representing the probability of the association rule “the data is included in the second data group DA2 in the analysis data group DA00 ′”.

  The data dividing unit 115 corresponds to the factor-undiscovered data extracting unit 10 in the data analysis apparatus according to the second embodiment, and the above-described division rule evaluation is selected from the input attribute conditions determined by the input attribute condition determining unit 111. The input attribute condition that maximizes the value is extracted, and the analysis data group DA00 ′ is divided into a factor data group that satisfies the input attribute condition and another data group that does not satisfy the input attribute condition.

  The factor determination unit 117 selects only an input attribute condition having a high priority from among a plurality of input attribute conditions related to the same input attribute extracted by the repetition process of the first factor extraction unit 109. .

  The complex factor defect count calculation unit 118 calculates the number of defects due to the complex factor of two conditions among the input attribute conditions selected by the factor determination unit 117.

  The numerical value-character data conversion unit 119 converts information representing the determined factor, for example, the numerical value of the input attribute threshold value xj-th in the determination factor list table and the composite factor table described later into character data.

  Next, the data analysis method and the input attribute condition determination method of the present embodiment will be described with reference to FIGS. 14 and 15 by taking the data group DA of Table 1 as a basic data group as an example. FIG. 14 is a flowchart showing the data analysis method of this embodiment, and FIG. 15 is a flowchart showing the input attribute condition determination method of this embodiment corresponding to the processing of step 107 of FIG. 14 (described later). is there.

  The basic data group DA in Table 1 is stored in the basic data group storage unit 102 such as a hard disk, and is composed of 12 pieces of data having ids (identifiers) of 1 to 12. In Table 1, x1, x2, x3, and x4 are input attributes. The input attribute x1 is a character attribute that takes one of four characters A, B, C, and D. The input attribute x2 is a character attribute that takes one of the four characters a, b, c, and d. The input attribute x3 is a discrete attribute that takes one of four discrete values 1, 2, 3, and 4. The input attribute x4 is a discrete attribute taking any one of four discrete values 10, 20, 30, and 40. The input attribute may be any of a character attribute, a discrete numerical attribute, and a continuous numerical attribute.

  In Table 1, y is an output attribute. The output attribute may be any of a character attribute, a discrete numerical attribute, and a continuous numerical attribute. Here, the output attribute is a character attribute that takes one of the three characters X, Y, and Z.

  In the data analysis method of the present embodiment, the case where y = Y is regarded as a problem event, and the cause of the output attribute y being Y is analyzed.

  As an example of the basic data group DA, for example, the input attribute is the manufacturing process condition and / or in-line inspection result (inspection result during the manufacturing line) in the product manufacturing process, the output attribute is the product quality determination result, Data in which the problem event y = Y is a bad quality determination result is exemplified. In this case, by analyzing the causal relationship between the input attribute and the output attribute by the data analysis method of this embodiment and deriving the cause of the problem event y = Y, it is possible to easily take measures to eliminate the occurrence of defective products. Is possible. Therefore, it is possible to easily improve the manufacturing process such as improvement in yield.

  As a more specific example of the basic data group DA, for example, input attributes x1, x2, x3, and x4 are process data such as gas flow rate, gas pressure, input power, and film formation time of plasma CVD process, and output. Data in which the attribute y is the film thickness of the thin film to be formed can be mentioned. The values of the input attribute and the output attribute may be any of a continuous numerical attribute, a discrete numerical attribute, and a character attribute. In the case of the character attribute, for example, the output attribute is an example of the film thickness, and is expressed as “large”, “medium”, and “small”.

[Step 100]
First, the character-numeric data conversion unit 1 converts the character attributes in the basic data group DA of Table 1 stored in the basic data group storage unit 102 such as a hard disk into numeric attributes (numeric data) according to the following conversion rules. (S100). The processing at step 100 is the same as the processing at step 0 in the data analysis method of the second embodiment.
(X1) A → 1, B → 2, C → 3, D → 4
(X2) a → 1, b → 2, c → 3, d → 4
(X3) No conversion (x4) No conversion (y) X → 1, Y → 2, Z → 3
Note that this processing is omitted when the input attribute and output attribute of the basic data group DA are originally numerical attributes. Therefore, when the input attribute and output attribute of the basic data group DA are originally numeric attributes, the character-numeric data conversion unit 1 can be omitted.

  Through the above processing, each data is converted into numerical data. Then, the character-numeric data conversion unit 1 sends the converted data group DA0 to the data classification unit 104.

  Here, it is preferable that the above conversion rule is set so that the numerical value of the output attribute increases as the input attribute value after conversion becomes larger as much as possible, or vice versa. There is no limitation to the above example. The data group DA0 converted into numerical data by the conversion rule is as shown in Table 2.

  The data group DA0 obtained by this conversion is a set of data composed of a plurality of input attributes (description attributes) and output attributes (target attributes), each consisting of a numerical attribute. Hereinafter, the data group DA0 is also referred to as a basic data group.

[Step 101]
The classification condition setting unit 103 performs basic data group DA in accordance with predetermined setting information or in response to a user inputting a problem event attribute value y = Y from an input unit such as a keyboard or a mouse (not shown). The classification condition of the output attribute y of the basic data group DA0 corresponding to the problem event y = Y is set and output to the data classification unit 104 (S101). In this example, the classification condition of the output attribute y of the basic data group DA0 corresponding to the problem event y = Y of the basic data group DA is y = 2. The processing in step 101 is the same as the processing in step 1 in the data analysis method of the second embodiment, except that the classification condition of the output attribute y is set instead of the threshold y _{th of the} output attribute y.

[Step 102]
Next, based on the value of the output attribute y of the basic data group DA0 and the classification condition (the following comparison logic (1) (2)) output from the classification condition setting unit 103, the data classification unit 104 The data group DA0 is classified into a first data group DA1 and a second data group DA2 (S102).

(1) y ≠ 2 → DA1
(2) y = 2 → DA2
In this case, the comparison logic (1), that is, “y ≠ 2” is a classification condition corresponding to an event of y ≠ Y in the basic data group DA (an event that is not a problem event; hereinafter referred to as “non-problem event”). The comparison logic (2), that is, “y = 2” is the classification condition corresponding to the problem event y = Y (hereinafter simply referred to as “problem event”) in the basic data group DA.

  Then, as shown in Table 3, a classification flag (“DA1” or “DA2”) corresponding to each data group is assigned (S102). Hereinafter, the data group in Table 3 is referred to as a post-classification basic data group DA00.

  The post-classification basic data group DA00 is stored in the post-classification basic data group storage unit 105 such as a hard disk.

  Here, the second data group DA2 is a data group representing a problem event (for example, a device characteristic failure or the like). That is, the second data group DA2 is a data group whose output attribute y is an attribute value (2) representing a problem event, and the first data group DA1 is an attribute value (1 or 3) whose output attribute y does not represent a problem event. ).

  The processing in this step 102 classifies the basic data group DA0 into the first data group DA1 and the second data group DA2, and then divides the basic data group DA0 into the first data group DA1 and the second data group DA2. This is the same as the processing in step 2 in the data analysis method of the second embodiment, except that the classification flags corresponding to the first data group DA1 and the second data group DA2 are assigned.

The classification by the data classification unit 104 is not limited to the above logic, and may be performed based on the logic based on the output attribute threshold y _th , for example, the following logic (1 ′) (2 ′).

(1 ') y> y _th → DA1
(2 ′) y ≦ y _th → DA2
In this case, the comparison logic (1 ′), that is, “y> y _th ” is the classification condition corresponding to the non-problem event, and the comparison logic (2 ′), that is, “y ≦ y _th ” corresponds to the problem event. This is a classification condition.

  Further, the classification by the data classification unit 104 may be performed based on the logic based on the logical sum or logical product of a plurality of conditions, for example, the following logic (1 ″) (2 ″).

(1 ″) y _th1 <y ≦ y _th2 → DA1
(Y _th1 <y AND y ≦ y _th2 )
Here, y _{th1 and} y _th2 are output attribute threshold values satisfying y _th1 <y _th2 (2 ″) y ≦ y _th1 OR y> y _th2 → DA2
In this case, the comparison logic (1 ″), that is, “y _th1 <y ≦ y _th2 ” is the classification condition corresponding to the non-problem event, and the comparison logic (2 ″), that is, “y ≦ y _th1 OR y > Y _th2 ”is the classification condition corresponding to the problem event.

Further, when the basic data group DA0 includes a plurality of output attributes (for example, results of a plurality of types of inspections), the classification by the data classification unit 104 is performed on the classification condition setting unit 103 for each of the output attributes y ₁ and y ₂ . It may be performed based on the logic based on the logical sum or logical product of the plurality of classification conditions set in (1), for example, the following logic (1 ′ ″) (2 ′ ″).

(1 ″ ′) y ₁ ≦ y _th1 OR y ₂ > y _th2 → DA1
(2 ″ ′) y ₁ > y _th1 AND y ₂ ≦ y _th2 → DA2
In this case, the comparison logic (1 ″), that is, “y ₁ ≦ y _th1 OR y ₂ > y _th2 ” is the classification condition corresponding to the non-problem event, and the comparison logic (2 ″), that is, “y ₁ > Y _th1 AND y ₂ ≦ y _th2 ”is the classification condition corresponding to the problem event.

Further, when there are a plurality of output attributes, the classification by the data classification unit 104 is a logic for one output attribute y selected from the plurality of output attributes y ₁ and y ₂ , for example, the logic (1) (2), You may perform based on the said logic (1 ') (2'), the said logic (1 '') (2 ''), etc.

[Steps 103-105]
The analysis data group extraction unit 106 extracts the analysis data group DA00 ′ to be analyzed from the classified basic data group DA00 and sends it to the data row separation unit 107.

  In this first process, the same data as the post-classification basic data group DA00 is extracted as the analysis data group DA00 '(S105). However, in the process of repeated processing (the second and subsequent processes) described later, the data dividing unit The other data group output by 115 is extracted (S104). That is, in steps 103 to 105, the analysis data group extraction unit 106 determines whether or not it is the first process (the analysis data group DA00 ′ has not been extracted so far) (S103). Based on the result, the basic data group DA00 after classification is extracted as the analysis data group DA00 ′ in the case of the first process, and the other data group is extracted as the analysis data group DA00 ′ in the case of the first process (S104 and S105).

[Step 106]
The data row separation unit 107 is based on each classification flag of the first data group DA1 and the second data group DA2 in the analysis data group DA00 ′ (basic data group DA00 after classification in the first process: Table 3). Then, the analysis data group DA00 ′ is divided into two, and each data group, that is, the first data group DA1 and the second data group DA2 are extracted (S106). Table 4 shows the first data group DA1 output from the data row separation unit 107, and Table 5 shows the second data group DA2.

  In the following description, the first data group DA1 is appropriately referred to as a non-defective product (OK product) data group, and the second data group DA2 is referred to as a defective product (NG product) data group.

  The above steps 102 to 105 replace step 2 in the data analysis method of the second embodiment.

[Step 107]
Next, the input attribute condition determining apparatus 100A specifically satisfies the input attribute condition so that the good product data group DA1 (first data group) and the defective product data group DA2 (second data group) are collected. The analysis data so that the defective data group DA2 (second data group) is grouped (biased) in the data group, and the non-defective data group DA1 (first data group) is grouped (biased) in the data group that does not satisfy the input attribute condition. An input attribute condition, which is an input attribute condition for bisecting the group DA00 ′, is determined (S107). Step S107 for determining the input attribute condition includes steps S203 to S208 as shown in FIG.

[Step 203]
In step 203, first, the data string extraction unit 5 extracts each data string of the input attribute xj (1 ≦ j ≦ 4) from the non-defective product data group DA1 (Table 4) (S203). This data string is called a 1-xj data group.

  Similarly, the data string extraction unit 5 extracts each data string of the input attribute xj (1 ≦ j ≦ 4) from the defective product data group DA2 (Table 5) (S203). This data string is called a 2-xj data group.

  The 1-xj data group is shown in Tables 6 to 9, and the 2-xj data group is shown in Tables 10 to 13. The processing at step 203 is the same as the processing at step 3 in the data analysis method of the second embodiment.

[Step 204]
The frequency calculation unit 6 inputs each of the 1-xj data group extracted from the non-defective product data group DA1 in step 203 and each of the 2-xj data group extracted from the defective product data group DA2 in step 203 to the input attribute xj. The rows are rearranged in ascending order with the value of (reordering process 1). For each numerical value of the input attribute xj, 1-xj frequency cumulative% (first frequency) representing the ratio of the number of data whose input attribute xj is equal to or less than the numerical value in the non-defective product data group DA1, and defective product data In the group DA2, 2-xj frequency cumulative% (second frequency) representing the ratio of the number of data whose input attribute xj is equal to or less than the numerical value is calculated (S204).

  Here, Tables 14 to 17 in which the data groups in Tables 6 to 9 are rearranged in ascending order by the value of the input attribute xj are used, and the data of each row (id) is in a position higher than the position of the data in the table. The ratio of the number of data to the total number of data (= 8) in the first data group is calculated as 1-xj frequency cumulative%. Similarly, using Table 18 to Table 21 in which Table 10 to Table 13 are rearranged in ascending order by the value of the input attribute xj, the number of data in each row (id) is equal to or greater than the position of the data in the table. The ratio of the second data group to the total number of data (= 4) is calculated as 2-xj frequency cumulative%.

  The values of 1-xj frequency cumulative% and 2-xj frequency cumulative% calculated here are shown in Tables 14 to 21.

  Further, the frequency calculation unit 6 is a table of 1-xj data groups that are non-defective product data groups for which 1-xj frequency cumulative% is calculated, and 2-items that are defective product data groups for which 2-xj frequency cumulative% is calculated. The table of the xj data group is joined (joining process). Specifically, for the input attribute x1, Table 14 and Table 18 are combined to generate the x1 frequency accumulation table in Table 22, and for the input attribute x2, Table 15 and Table 19 are combined to generate the x2 frequency accumulation in Table 23. For the input attribute x3, the table 16 and the table 20 are combined to generate the x3 frequency accumulation table in Table 24, and for the input attribute x4, the table 17 and the table 21 are combined to generate the x4 frequency accumulation table in the table 25. , Respectively (S204).

  Further, the frequency calculation unit 6 rearranges the rows in the frequency accumulation tables of Tables 22 to 25 in ascending order by the value of the input attribute xj (sorting process 2). After the rearrangement process 2, the value immediately above (the value of the data on one line) is assigned to the blanks of 1-xj frequency accumulation% and 2-xj frequency accumulation% in order from the upper blank (substitution process). . Thereafter, only the data of the last line among the lines with the same value in the input attribute xj is adopted (duplicate processing). In this way, as shown in Tables 26 to 29, the ratio of the number of data whose input attribute xj is equal to or less than the numerical value in the first data group DA1 which is a non-defective data group is represented for each value of the input attribute xj. 1-xj frequency accumulation% (A; first frequency) and 2-xj frequency accumulation representing the ratio of the number of data whose input attribute xj is less than or equal to the numerical value in the second data group DA2, which is a defective product data group. % (B; second frequency) is calculated (S204). The processing in step 204 is the same as the processing in step 4 in the data analysis method of the second embodiment.

  In steps 203 and 204 described above, the xj frequency accumulation table of Table 26 to Table 29 is created in order to create a data string extraction process (Table 6 to Table 13) → rearrangement process 1 → 1−xj frequency accumulation% and 2 -Xj frequency cumulative% calculation processing (Table 14 to Table 21) → join processing (Table 22 to Table 25) → sort processing 2 → assignment processing → duplication processing (Table 26 to Table 29) The calculation may be performed so as to directly create the xj frequency accumulation table of Table 26 to Table 29 without performing the individual processing.

[Step 205]
Next, the frequency accumulation difference calculation unit 7 calculates, for each value of the input attribute xj, a difference between 1-xj frequency accumulation% (A) of non-defective product and 2-xj frequency accumulation% (B) of defective product (= | A−B |) is calculated (S205). This difference value is referred to as xj frequency cumulative difference%. Tables 30 to 33 show the calculation results of the xj frequency cumulative difference%.

  FIG. 3 shows the relationship between the value of the input attribute xj and the non-defective product 1-xj frequency cumulative% (A), the defective product 2-xj frequency cumulative% (B), and the xj frequency cumulative difference% | A-B |. To FIG.

  The xj frequency cumulative difference% with respect to each numerical value of the input attribute xj is divided into the first data group DA1 of defective products and the defective products by dividing into the range where the input attribute xj is less than the numerical value and the range where the input attribute xj exceeds the numerical value. Is a threshold evaluation index indicating whether or not the second data group DA2 is well separated, and corresponds to the improvement degree of the Gini index method.

  That is, the xj frequency cumulative difference% (= | A−B |) in each numerical value of the input attribute xj is “data belonging to the second data group of defective products if the input attribute xj is equal to or smaller than the numerical value. A correlation rule that “if xj exceeds the numerical value, the data belongs to the first data group of non-defective products” or “if the input attribute xj exceeds the numerical value, the data belongs to the second data group of defective products. Yes, if the input attribute xj is less than or equal to the numerical value, it indicates the probability of the correlation rule that the data belongs to the first good data group.

  The “relationship between the value of the input attribute xj and the xj frequency cumulative difference% | A−B |” (FIGS. 3 to 6) according to the present embodiment is “the branch condition of the input attribute xj” in the Gini index method. , “Relationship with degree of improvement” (FIGS. 20 to 23).

  The processing in step 205 is the same as the processing in step 5 in the data analysis method of the second embodiment.

[Steps 206 to 208]
The threshold value determination unit 130 extracts, for each input attribute xj, the value of the input attribute xj having the maximum value among the xj frequency cumulative difference% with respect to each value of the input attribute xj (S206). The processing in step 206 is the same as the processing in step 6 in the data analysis method of the second embodiment.

  In Tables 30 to 33, the extracted values are shown in gray. The value of the extracted input attribute xj will be referred to as an input attribute threshold value xj-th. As can be seen with reference to FIGS. 3 to 6, the input attribute threshold value xj-th is divided into a range of xj ≦ xj-th and a range of xj> xj-th, so that the non-defective first data group DA1. And the value of the input attribute xj that makes it easy to separate the defective product from the second data group DA2.

  Next, the polarity determination unit 131 determines the magnitude relationship between the 1-xj frequency cumulative% (A) of the non-defective product and the 2-xj frequency cumulative% (B) of the defective product at the threshold value xj-th of each input attribute xj. Determination is made (S207). In Tables 30 to 33, the type of xj frequency accumulation (1-xj frequency accumulation or 2-xj frequency accumulation) determined to be larger in the threshold value xj-th is also shown.

  Next, based on the threshold value xj−th determined (extracted) by the threshold value determination unit 130 and the xj frequency accumulation type determined to be large by the polarity determination unit 131, the input attribute condition determination unit 111 determines that the defective product is defective. The input attribute condition corresponding to the second data group DA2 of the above, that is, “if the input attribute satisfies the input attribute condition, the data belongs to the second data group of the defective product, and if the input attribute does not satisfy the input attribute condition, An input attribute condition that satisfies a specific correlation rule that the data belongs to the first data group is determined (S208).

  If the xj frequency accumulation type determined to be large by the polarity determination unit 131 is the non-defective 1-xj frequency accumulation type, the input attribute condition “xj> xj-th” is the number of defective products. When the type of xj frequency accumulation that is determined to be large by the polarity determination unit 131 corresponding to the two data groups DA2 is 2-xj frequency accumulation of defective products, an input “xj ≦ xj−th” is input. The attribute condition corresponds to the defective second data group DA2. Therefore, when the polarity determination unit 131 determines that the 1-xj frequency cumulative% (A) is larger, the input attribute condition determination unit 111 determines the input attribute corresponding to the second data group DA2 of the defective product. If the condition is determined as “xj> xj−th (input attribute exceeds the threshold)” and the polarity determination unit 131 determines that 2-xj frequency cumulative% (B) is larger, it is not acceptable. The input attribute condition corresponding to the non-defective second data group DA2 is determined to be a condition of “xj ≦ xj−th (input attribute is equal to or less than threshold)”.

  Table 34 shows the input attribute conditions determined by the input attribute condition determination unit 111 as conditions corresponding to the defective second data group DA2 as described above.

  As an example, an input attribute condition “x2> 2” is determined for the input attribute x2. This input attribute condition indicates a condition that can be separated from the non-defective first data group DA1 to detect the defective second data group DA2 with high accuracy. Also, “x2 ≦ 2”, which is an exclusive condition for the determined input attribute condition “x2> 2”, separates the defective second data group DA2 from the defective first data group DA1 with high accuracy. The conditions that can be detected are shown. These can be better understood with reference to FIG.

  In the above description, the processing from step 203 to step 208 is collectively performed for a plurality of input attributes. However, the processing from step 203 to step 208 may be repeated by sequentially increasing the value of j from 1 to 4. .

  The processing from step 203 to step 208 corresponds to the input attribute condition determination method in the claims. In this input attribute condition determination method of this embodiment, it is very simple to calculate only the xj frequency cumulative difference%, which is the difference between the non-defective 1-xj frequency cumulative% and the defective product 2-xj frequency cumulative%. In the processing, a threshold evaluation index (corresponding to the improvement degree of the Gini index method) is obtained. That is, as in the Gini index method, the Gini index (formula (3) (4)) and the improvement level (formula (6)) are calculated for every branch condition pattern that the input attribute can take. Without performing the arithmetic processing, only the arithmetic processing for obtaining the frequency cumulative difference is performed on the data of the number of rows corresponding to the number of values that the input attribute can take (Tables 30 to 33). Therefore, even if the number of values that can be taken by one input attribute is on the order of tens of thousands to hundreds of thousands as in the data of the manufacturing process of a product such as an actual device (especially a semiconductor device), Table 30 to Since only the number of data rows in Table 33 is increased, almost no calculation load is applied, and processing can be performed in a short time.

  Further, the 1-xj frequency cumulative% of non-defective products and the 2-xj frequency cumulative% of defective products for each value of the input attribute xj respectively indicate the number of data whose input attributes are equal to or less than the numerical value in the corresponding data group. Since the data is normalized by the total number of data in the data group, the xj frequency cumulative difference% (threshold evaluation index) that is the difference between these is the ratio of non-defective products (first data group) in the analysis data group. Even if the ratio of non-defective products (second data group) is extremely different, the input attribute condition (optimal branch condition for each input attribute) that separates non-defective products and defective products is high without reducing the accuracy. Can be determined with accuracy. In the above example using the data group in Table 1, the ratio of defective products is 4/12 and the ratio of non-defective products is 8/12, and there is no extreme (digit difference) difference between the two. Therefore, the condition of the defective product extracted in the present embodiment (Table 34) matches the branch condition extracted by the Gini index method (the condition described in [Problems to be solved by the invention]).

  As described above, according to the input attribute condition determination method of the present embodiment, both the first object and the second object of the present invention can be achieved.

  The processing after Step 109 is a suitable data analysis method utilizing the input attribute condition determined by the above-described input attribute condition determination method, and the processing content will be described below.

[Step 109]
For each of the input attribute conditions (Table 34) determined by the input attribute condition determination unit 111, the defective product separation degree calculation unit 112 includes the post-classification basic data group DA00 (not the analysis data group DA00 ′). The number of data satisfying the input attribute condition ("DA1 + DA2" column in Table 35) and the number of data satisfying the input attribute condition and corresponding to the defective second data group DA2 ("DA2" column in Table 35) ). Then, a defective product separation degree is calculated by dividing the value in the “DA2” column of Table 35 by the value in the “DA1 + DA2” column (S109). The defective product separation degree of each input attribute condition represents the accuracy of defective product extraction based on the input attribute condition (defective rate when data belonging to the input attribute condition in the basic data group DA00 after classification is used as a population). This corresponds to the second data group separation degree in the claims.

  Table 35 shows the total number of data (value in the “DA1 + DA2” column = 12) in the basic data group DA00 after classification in the “Total” row, together with the calculation result by the defective product separation degree calculation unit 112. A table showing the number of non-defective second data groups DA2 (value of “DA2” column = 4) and defective product content (value of “defective product separation” column = 4/12 = 0.333) is there. The defective product content rate represents the defective rate when all the data of the post-classification basic data group DA00 is a population, and corresponds to the second data group content rate in the claims.

  The meaning of each column in Table 35 can be easily understood with reference to FIG. The meaning of each column in Table 35 can be easily understood by referring to FIG. 16A to FIG. FIGS. 16A to 16D are Venn diagrams showing the relationship between a set of data satisfying the input attribute condition of each column of Table 35 and a set of defective second data group DA2.

[Step 110]
The first factor extraction unit 109 includes, in each of the input attribute conditions in Table 35, the defective product content rate in the basic data group DA00 after classification (the value of the “defective product separation” column in the “Total” row = 0. 333) is extracted as information indicating the factor of the second data group DA2 of defective products. Then, the result is stored in the analysis result data storage unit 14.

  In the example of Table 35, since all the input attribute conditions for x1 to x4 have a defective product separation degree higher than the defective product content rate of the basic data group DA00 after classification, all input attribute conditions are extracted. (Table 36).

  Each input attribute condition in Table 36 is a condition that includes the second data group DA2 of defective products at a higher rate than the sample randomly selected from the basic data group DA00 after classification, and the second data group of defective products. The cause of the output attribute condition corresponding to the data group is shown.

  As described above, the input attribute conditions “x1> 2,” “x2> 2,” “x3> 2,” “x4 ≦ 10” are caused as the cause of the problem event (defective product second data group DA2). Extracted.

  In steps 101 to 110 described above, the cause of the problem event (second data group DA2 of defective products) can be extracted.

  However, in the process (step 204), the threshold evaluation index (xj frequency cumulative difference%) calculated for each value taken by each input attribute includes the influence of the input attribute other than the input attribute as a disturbance. In some cases, the accuracy of the threshold evaluation index (xj frequency cumulative difference%) may be reduced. Further, when the cause of the problem event should be two types of “xj ≦ xj−th1” and “xj> xj−th2” in a certain input attribute xj, the processing of steps 203 to 208 alone Only one of the factors is extracted. In order to eliminate these points, it is preferable to further perform processing according to the following steps.

[Step 111]
For each input attribute threshold value xj-th (see Tables 30 to 33) determined (extracted) by the threshold value determination unit 130 (step 206) by the frequency accumulation ratio calculation unit 16, 1-xj frequency accumulation percentage (A) of non-defective products The ratio of 2-xj frequency cumulative% (B) of defective products to (= B / A: hereinafter referred to as the lower frequency cumulative ratio) or 100 minus 1-xj frequency cumulative% (A) of non-defective products Ratio of value (= 100-B) obtained by subtracting 2-xj frequency cumulative percentage (B) of defective products from 100 to value (= 100-A) (= (100-B) / (100-A): below , Referred to as frequency cumulative upper ratio) as a division rule evaluation value.

  If the input attribute condition determined by the input attribute condition determination unit 111 is a type of “xj ≦ xj−th” (the type of xj frequency accumulation that is determined to be large by the polarity determination unit 131 is In the case of 2-xj frequency accumulation of defective products), the frequency accumulation lower ratio (= B / A) is calculated as the division rule evaluation value. Here, the frequency cumulative lower ratio (= B / A) is a ratio at which the second data group of defective products can be detected separately from the first data group of non-defective products according to the input attribute condition “xj ≦ xj−th”. Represents.

  Further, when the input attribute condition determined by the input attribute condition determination unit 111 is a type of “xj> xj−th” (the type of xj frequency accumulation that is determined to be large by the polarity determination unit 131 is If the non-defective product is 1-xj frequency cumulative), the frequency cumulative upper ratio (= (100-B) / (100-A)) is calculated as the division rule evaluation value. Here, the cumulative frequency ratio (= (100−B) / (100−A)) is separated from the non-defective first data group by the input attribute condition “xj> xj−th”. This represents the ratio at which two data groups can be detected.

  In other words, the division rule evaluation value (frequency cumulative lower ratio or frequency cumulative upper ratio) is “for the second data group in the analysis data group if the input attribute xj satisfies the input attribute condition” for each input attribute condition. It represents the certainty of the association rule that it is “included data”.

  Table 37 shows the division rule evaluation values (frequency cumulative lower ratio or frequency cumulative upper ratio) for each input attribute condition.

  Note that the processing in step 111 does not calculate two types of rule evaluation values (first and second rule evaluation values), but the type of input attribute condition determined by the input attribute condition determination unit 111 ( Data analysis according to the second embodiment except that the frequency cumulative lower ratio or the frequency cumulative upper ratio is calculated as the division rule evaluation value according to the xj frequency cumulative type determined to be larger by the polarity determination unit 131) This is similar to the processing in step 7 of the method.

[Step 112]
Next, the data dividing unit 115 selects the value of the division rule evaluation value (frequency cumulative lower ratio or frequency cumulative upper ratio; Table 37) in step 111 from the input attribute conditions determined by the input attribute condition determining unit 111. An input attribute condition that maximizes is extracted (S112).

  Referring to Table 37, the input attribute condition “x2> 2” has the largest division rule evaluation value among all the input attribute conditions, and division rule evaluation value = frequency cumulative upper ratio = ∞. This indicates that, under the input attribute condition “x2> 2,” the defective second data group DA2 can be detected by being completely separated from the non-defective first data group DA1.

  From another point of view, the input attribute condition “x2> 2” corresponds to the defective second data group DA2 regardless of the values of the other input attributes (x1, x3, x4). From the above, when determining the input attribute conditions of the other input attributes (x1, x3, x4) (steps 203 to 206) or calculating the threshold evaluation index (xj frequency cumulative difference%) (step 205) May be a disturbance factor. In such a case, it is desirable to obtain the input attribute conditions of the other input attributes (x1, x3, x4) by excluding data corresponding to “x2> 2” from the analysis data group DA00 ′.

  Therefore, the data dividing unit 115 satisfies the analysis data group DA00 ′ based on the extracted input attribute condition “x2> 2”, the factor data group satisfying “x2> 2”, and “x2> 2”. It is divided into other data groups which do not exist (satisfying “x2 ≦ 2”) Table 38 shows the factor data group, and Table 39 shows the other data group.

  This step 112 corresponds to step 9 in the data analysis method of the second embodiment. Here, the frequency cumulative lower ratio or the frequency cumulative upper ratio is calculated as the division rule evaluation value, but other evaluation indexes such as the Gini index and the above-described frequency cumulative difference may be used.

[Step 113]
Next, the analysis data group extraction unit 106 extracts another data group as the next analysis data group DA00 ′ from the data group divided in step 112 (S104). Then, the processing from step 106 to step 113 is repeated until the end condition determination unit 11 determines that the end condition is satisfied. That is, after the second step 106, the process proceeds to step 113, where it is determined whether or not the end condition is satisfied by the end condition determining unit 11 (S113). If the end condition determination unit 11 determines that the end condition is not satisfied, the processing of steps 107 to 112 and step 104 is performed again, and the end condition determination unit 11 determines that the end condition is satisfied. If so, the process proceeds to step 114. The determination of the end condition in step 113 is the same as the determination of the end condition in step 10 in the data analysis method of the second embodiment.

  The end condition determination unit 11 of the present embodiment is configured to determine the end condition when the number of data in the second data group DA2 of defective products becomes 0 in step 106 during the repetitive processing. In this way, by repeatedly performing the process until the number of data in the second data group DA2 of defective products becomes zero, a detailed factor analysis result for the second data group DA2 of defective products can be obtained.

  Note that the end condition is another end condition based on the number of data in the second data group DA2, for example, (1) when the number of data in the second data group DA2 is equal to or less than a predetermined number in step 106 during the repetitive processing. (2) When the ratio of the number of data in the second data group DA2 to the number of data in the first data group DA1 is equal to or less than a predetermined ratio in the step 106 during the iterative processing, (3) step 112 during the iterative processing. The division rule evaluation value of the input attribute condition extracted in (1) may be lower than a predetermined value. When such termination conditions are used, a simpler and sufficient factor analysis result can be obtained. Further, when priority is given to obtaining a concise factor analysis result, the end condition is simply a case where the iterative process is performed a predetermined number of times, or the end condition determining unit 11 is omitted and the iterative process is performed as much as possible. May be.

  Table 40 (corresponding to Table 35 for the first time) shows the calculation results of the defective product separation degree calculation unit 112 in Step 109 for the second time.

  In the example of Table 40, the input attribute condition of x1, x3, and x4 has a defective product separation degree (second data group separation degree) higher than the defective product content rate (second data group content rate) of the basic data group DA00 after classification. These conditions are extracted (second step 110; Table 41).

  On the other hand, since the defective product separation degree (second data group separation degree) under the input attribute condition (x2 ≦ 1) of x2 is lower than the defective product content rate of the basic data group DA00 after classification, the input attribute of x2 Condition is not extracted.

  As described above, the problem event (the second data group DA2 of defective products) is obtained by the second processing (processing in which the data satisfying “x2 ≦ 2” in the basic data group DA00 after classification is set as the analysis data group). As input factors, input attribute conditions of “x1> 2”, “x3> 2”, and “x4 ≦ 10” were extracted (Table 41).

  The division rule evaluation values (frequency cumulative lower ratio or frequency cumulative upper ratio) calculated in the second step 111 are shown in Table 42 (corresponding to the first table 37).

  In this example, the division rule evaluation value is a maximum of 4 in “x1> 2” and “x4 ≦ 10”, but the data division unit 115 selects one of these. This selection criterion should have a certain rule. For example, the input attribute xj with a smaller number j is given priority, and “x1> 2” is selected (second step 112).

  Of the data group divided by the data dividing unit 115, another data group (a data group satisfying “x1 ≦ 2” in the second analysis data group) is the analysis data group extraction unit as the third analysis data group. (Table 43).

  However, since the analysis data group for the third time in Table 43 did not include defective product data (second data group DA2; y = 2), it is repeated at this point (up to the second factor extraction). Processing has ended.

[Step 114]
Table 44 shows an extraction factor list table in which the input attribute conditions (Table 35 and Table 41) extracted for each repetition of Step 110 are summarized.

  The extracted factor list table of Table 44 shows all of the plurality of input attribute conditions for the same input attribute by the repetition processing of the first factor extracting unit 109 (step 110).

  The factor determination unit 117 selects only an input attribute condition having a high priority among a plurality of input attribute conditions (Table 44) for the same input attribute (S114).

  Specifically, for the same input attribute, the input attribute condition that maximizes the defective product separation degree (second data group separation degree) in the first pattern “input attribute is below threshold” And one input attribute condition that maximizes the defective product separation degree (second data group separation degree) in the second pattern of “input attribute exceeds threshold”.

  In the case of the example in Table 44, finally, the four conditions shown in Table 45 were selected as factors of the problem event (defective product second data group DA2).

  Table 45 is a list of input attribute conditions determined (selected) by the factor determination unit 117 as the cause of the problem event, and this table is referred to as a determination factor list table. The determination factor list table is stored in the analysis result data storage unit 14.

  If the process of the above two iterations is expressed in the form of a decision tree (the same format as in FIG. 12), it is as shown in FIG. Referring to FIG. 17, in this embodiment, every decision tree branch causes a problem event (defective second data group DA2) for all input attributes, not just the final branch condition. An input attribute condition is obtained (processing by the input attribute condition determination unit 111 in step 208), and only input attribute conditions having a high degree of defective product separation are extracted (by the first factor extraction unit 109 in step 110). processing). Then, among all the input attribute conditions for the number of times of branching (for the number of iterations), input attribute conditions with a higher degree of defective product separation are narrowed down and determined as the final defective factor (factor of step 114) Processing by the determination unit 117).

  In this way, even if it is a condition other than the branch condition in the decision tree, all the conditions with a high degree of defective product separation are extracted, so even if there are competing factors in the branch condition, without missing that factor, I can capture it reliably. In addition, factor extraction for each branch (processing by the first factor extraction unit 109) and final factor determination (processing by the factor determination unit 117) are performed based on a clear index of defective product separation. Or, since the decision is made, the cause of the problem event can be clearly grasped no matter how complicated the decision tree is. Furthermore, since the degree of defective product separation is used as an evaluation index, it is possible to prioritize a plurality of determined factors (input attribute conditions).

[Step 115]
The complex factor defect count calculation unit 118 calculates the number of defects due to the complex factor of two conditions among the input attribute conditions in the decision factor list table (Table 45) (Table 46).

In Table 46, the title row and the title column indicate the input attribute conditions of the decision factor list table, respectively, and the number of defects (second data) due to the composite factor of the two input attribute conditions is shown at the intersection. The number of groups DA2) is shown. For example, “x1> 2” row and “x2> 2” column are
This represents the number of data (= 1) satisfying “x1> 2” and “x2> 2” and corresponding to the defective second data group DA2. Hereinafter, the table in Table 46 is referred to as a composite factor table.

[Step 116]
The numerical value-character data conversion unit 119 converts the numerical value of the input attribute threshold value xj-th in the determination factor list table (Table 45) and the composite factor table (Table 46) into character data as necessary. The conversion rule to character data is a rule that is the reverse conversion of the conversion in step 100, and is as follows.
(X1) 1 → A, 2 → B, 3 → C, 4 → D
(X2) 1 → a, 2 → b, 3 → c, 4 → d
(X3) Not converted (x4) Not converted Table 47 shows a factor list table in which the input attribute threshold value xj-th in the determination factor list table in Table 45 is converted into character data.

[Step 117]
The data analysis is completed as described above, and the extraction factor list table (Table 44), the decision factor list tables (Table 45, Table 47), the composite factor table (Table 46) and various information in the data analysis process are finally The analysis result data is stored in the analysis result data storage unit 14 such as a hard disk. The analysis result data is appropriately sent from the analysis result data storage unit 14 to the output unit 15 such as a display device or a printing device, and displayed as a table (for example, Table 47), a decision tree (for example, FIG. 17), or a graph. It can be displayed on the device or printed on the printing device.

  As an example, FIG. 18 shows an example in which the decision factor list table (Table 47) is displayed as a factor breakdown Pareto diagram. In FIG. 18, the number of defects (number of second data groups DA2) due to each input attribute condition in the determination factor list table (Table 47) is represented by a bar graph, and the degree of defective product separation (second data group separation degree) is shown. This is shown by a line graph.

  By referring to the result of FIG. 18, the user can immediately determine the cause of the product characteristic failure such as “Which value range is each of the input attributes x1 to x4? it can. Further, the order (priority order) in which measures should be taken can be determined from the defective product separation degree (second data group separation degree). Furthermore, as a result of the countermeasure against the input attribute condition of FIG. 18, it can be estimated from the number of defects (number of second data groups DA2) how much the number of defects can be reduced.

  In the case of the example in FIG. 18, the first countermeasure is the input attribute x2 (“x2> 2”, that is, “x2 = c or d”) having the highest defective product separation degree (second data group separation degree). This measure should eliminate two of the four defects (50% of the total defects are resolved).

  As for the content to be secondly countermeasured, a composite factor table (Table 46) is used to combine the first factor (“x2> 2”, ie, “x2 = c or d”) with other factors. It can be judged by examining the degree. FIG. 19 shows the number of defects due to a composite factor with the first factor (“x2> 2”, that is, “x2 = c or d”) in the bar graph (number of defects) of each factor (input attribute condition) in FIG. It is shown with hatching. From FIG. 19, “x1> 2”, that is, “x1 = C or D”) has a high degree of defective product separation (second data group separation) and the first factor (“x2> 2”, ie, Since there are many defects that do not overlap with “x2 = C or D”), there is a high possibility of an independent factor with respect to the first factor, and it can be read that this is an item to be secondly countered.

  In FIG. 19, it is also possible to extract a composite factor (or dependent factor) with the first factor (“x2> 2”, ie, “x2 = C or D”). In this example, hatching is performed. “X4 ≦ 10” having a large part ratio is extracted.

  The data analysis method (step 100 to step 117) of the present embodiment has the following effects in addition to the operational effects of the above-described input attribute condition determination method (the processing of steps 203 to 208 (step 107)). .

  That is, it is clear that the defective product separation degree (second data group separation degree) indicates the accuracy of defective product extraction (the defective rate when the data belonging to the input attribute condition in the basic data group DA00 after classification is a population). Since the factors of defects are determined based on various indices, the priorities of the extracted factors can be prioritized, and the factor list in Table 47 (or Table 45) or the factor breakdown Pareto chart in FIG. The cause of the problem event can be derived in a very simple form as shown in. By using this, it is possible to obtain the defective product separation degree (second data group separation degree) and the number of defects of each factor (input attribute condition) for the problem event.

  In the above-described embodiment, a decision tree with a plurality of branches (repetition) is generated. However, if only one branch is required, the decision tree may be ended in step 110.

  In the above description, in step 113, the analysis data group extraction unit 106 extracts only the other data group from the divided data groups as the next analysis data group. However, the factor data group is also extracted as the analysis data group. The processing of step 106 to step 113 may be repeated. Thereby, a more detailed analysis can be performed.

  In the above description, the data analysis example in which the second data group DA2 is the defective data group and the cause of the defect is extracted has been described. However, the second data group DA2 is the good data group and the condition for obtaining the good product is shown. It is good also as data analysis which extracts.

  In the input attribute condition determination method and the data analysis method described above, the computer performs steps S203 to S208 (step 203 to step 208) and S100 to S117 (step 100 to step 117) in FIGS. This can be realized by executing a program including a corresponding process.

  Therefore, in the input attribute condition determining apparatus 100A in FIG. 13, the input attribute condition determining program includes a computer, a data row separating unit 107, a data string extracting unit 5, a frequency calculating unit 6, a frequency cumulative difference calculating unit 7, a threshold determination. This can be realized by functioning as the unit 130, the polarity determination unit 131, and the input attribute condition determination unit 111.

  In the data analysis apparatus 100 of FIG. 13, the data analysis program includes a computer, a character-numeric data conversion unit 1, a classification condition setting unit 103, a data classification unit 104, an analysis data group extraction unit 106, and a data row separation unit 107. , Data string extraction unit 5, frequency calculation unit 6, frequency cumulative difference calculation unit 7, threshold value determination unit 130, polarity determination unit 131, input attribute condition determination unit 111, defective product separation degree calculation unit 112, first factor extraction unit 109, the frequency accumulation ratio calculation unit 16, the data division unit 115, the end condition determination unit 11, the factor determination unit 117, the complex factor defect number calculation unit 118, and the numerical value-character data conversion unit 119 can be realized. It is.

  The program can be provided to the user by storing it in a computer-readable recording medium. The recording medium may be a built-in medium built in the computer main body, or a removable medium configured to be separable from the computer main body. Examples of the built-in medium include ROM; rewritable nonvolatile memory such as flash memory; and hard disk. In addition, as the removable media, optical recording media such as CD-ROM and DVD; magneto-optical recording media such as MO; magnetic recording media such as floppy (registered trademark) disk, cassette tape, and removable hard disk; memory cards and the like And a medium having a built-in rewritable nonvolatile memory such as a medium having a built-in ROM such as a ROM cassette.

  The program may be configured to be executed by CPU access, or after reading the program stored in the recording medium and transferring the read program to the program storage area of the built-in medium, the program on the built-in medium May be executed by CPU access. In addition, the program is not limited to be sold in a state where it is stored in a computer-readable recording medium, and is sold in a format that is transferred to a user's computer via a communication network such as the Internet. It may be a thing.

The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope shown in the claims, and embodiments obtained by appropriately combining technical means disclosed in different embodiments. Is also included in the technical scope of the present invention.
[Modification of Embodiment 1]
In the first embodiment, the input factor condition determined by the input attribute condition determining unit 111 by the first factor extracting unit 109 (step 110) is higher than the defective product content rate of the basic data group DA00 after classification. An input attribute condition having a large value of defective product separation is extracted as information indicating the factor of the second data group DA2 of defective products. That is, for each branch of the decision tree, not only the final branch condition, but also the input attribute conditions that cause the problem event (defective product second data group DA2) for all input attributes, and among these, An input attribute condition with a high degree of defective product separation was extracted.

  However, depending on the purpose of data analysis, simplicity may be required rather than such detailed analysis. In such a case, only the branch condition in the decision tree has to be extracted.

  In this modification, the input attribute condition (branch condition in the decision tree) having the maximum division rule evaluation value extracted by the data division unit 115 is extracted after the data division unit 115 of the first embodiment (FIG. 13). A second factor extraction unit (not shown) for extracting as an input attribute condition that causes a problem event (second data group DA2 of defective products) is provided. In this case, the first factor extraction unit 109 can be omitted.

[Embodiment 2]
Other embodiments of the present invention are described below. For convenience of explanation, members having the same functions as those shown in the first embodiment are denoted by the same reference numerals, and description thereof is omitted.

  First, the data analysis apparatus of this embodiment is demonstrated based on FIG.

  As shown in FIG. 1, the data analysis apparatus includes a character-numerical data conversion unit 1, an analysis target data storage unit 2, a threshold setting unit (threshold setting unit) 3, a data classification unit (classification unit) 4, and a data string extraction unit. 5, frequency calculation unit (frequency calculation unit in the first evaluation unit) 6, frequency cumulative difference calculation unit (difference calculation unit in the first evaluation unit) 7, input attribute threshold determination unit (threshold determination unit) 8. Frequency cumulative ratio calculation unit (second evaluation unit) 16, second factor extraction unit (second factor extraction unit) 9, factor undiscovered data extraction unit (division unit) 10, end condition determination unit (end) A condition determination unit) 11, an input attribute threshold value table creation unit 12, a contribution rate calculation unit 13, an analysis result data storage unit 14, and an output unit 15.

  Next, the data analysis method of the present embodiment will be described with reference to FIG. 2, taking as an example the case where the data group DA in the following Table 48 is an analysis target. The data group DA in Table 48 is stored in the analysis target data storage unit 2 such as a hard disk.

  The data group DA in Table 48 includes 12 pieces of data having ids (identifiers) 1 to 12. In Table 48, x1, x2, x3, and x4 are input attributes. The input attribute x1 is a character attribute that takes one of four characters A, B, C, and D. The input attribute x2 is a character attribute that takes one of the four characters a, b, c, and d. The input attribute x3 is a discrete attribute that takes one of four discrete values 1, 2, 3, and 4. The input attribute x4 is a discrete attribute taking any one of four discrete values 10, 20, 30, and 40. The input attribute may be a continuous attribute that takes a continuous numerical value.

  In Table 48, y is an output attribute. The output attribute may be a character attribute, a discrete attribute, or a continuous attribute. Here, the output attribute is a character attribute that takes one of three characters X, Y, and Z.

  In the data analysis method of the present embodiment, the case where y = Y is regarded as a problem event, and the cause of the output attribute y being Y is analyzed.

  As an example of the analysis target data, for example, the input attribute is the manufacturing process condition and / or in-line inspection result (inspection result during the manufacturing line) in the product manufacturing process, the output attribute is the product quality determination result, y Data in which the problem event = Y is a bad quality determination result can be cited. In this case, the causal relationship between the input attribute and the output attribute is analyzed by the data analysis method of the present embodiment, and the cause of the problem event y = Y is derived, thereby eliminating the occurrence of defective products such as device characteristic defects. Can be easily achieved. Therefore, it is possible to easily improve the manufacturing process such as improvement in yield.

  As a more specific example of analysis target data, for example, input attributes x1, x2, x3, and x4 are process data such as gas flow rate, gas pressure, input power, and film formation time of plasma CVD process, and output attributes. Data in which y is the thickness of a thin film formed by a plasma CVD process can be given. The values of the input attribute and output attribute may be continuous attributes, discrete attributes, or character attributes. In the case of the character attribute, for example, the output attribute is an example of the film thickness, and is expressed as “large”, “medium”, and “small”.

[Step 0]
First, the character-numeric data conversion unit 1 converts the character attributes in the data group DA of Table 48 stored in the analysis target data storage unit 2 such as a hard disk into numerical attributes (numeric data) according to the following conversion rule (S0). ). Thereby, each data is converted into numerical data. Then, the character-numeric data conversion unit 1 sends the converted data group to the data classification unit 4.
(X1) A → 1, B → 2, C → 3, D → 4
(X2) a → 1, b → 2, c → 3, d → 4
(X3) No conversion (x4) No conversion (y) X → 1, Y → 2, Z → 3
It is preferable that the conversion rule is set so that the numerical value of the output attribute increases as the numerical value of the input attribute after conversion increases, or vice versa. The conversion rule is not limited to the above example as long as it is unique.

  The data group DA0 converted into numerical data by the conversion rule is as shown in Table 49.

  By this conversion, the obtained data group DA0 becomes a set of data composed of a plurality of input attributes (description attributes) and output attributes (target attributes) that are numerical attributes. Hereinafter, the data group DA0 is referred to as a basic data group.

[Step 1]
The threshold value setting unit 3 responds to predetermined setting information or in response to the user inputting the problem event attribute value y = Y from an input unit such as a keyboard or a mouse (not shown). The threshold value (output attribute threshold value) y _th of the output attribute y of the basic data group DA0 corresponding to the problem event = Y is set and output to the data classification unit 4 (S1). In this example, the threshold y _th of the output attribute y of the basic data group DA0 corresponding to the problem event y = Y of the data group DA is y _th = 2.

[Step 2]
Next, the data classification unit 4 determines the basic data based on the comparison logic (1) (2) between the value of the output attribute y of the basic data group DA0 and the output attribute threshold y _th output from the threshold setting unit 3. The group DA0 is divided into two groups (classification) into a first data group DA1 and a second data group DA2 (S2).

(1) y> y _th or y <y _th → DA1
(2) y = y _th → DA2
In other words, the data classification unit 4 includes the basic data group DA0, the first data group DA1 whose output attribute does not match the output attribute threshold y _th (that is, 1 or 3), and the output attribute that is the output attribute threshold y _th ( = 2) and the second data group DA2 that coincides with the second data group DA2. The second data group DA2 is a data group of problem events (for example, defective device characteristics). That is, the second data group DA2 is a data group whose output attribute y is an attribute value (2) representing a problem event, and the first data group DA1 is an attribute value (1 or 3) whose output attribute y does not represent a problem event. ).

  Table 50 shows the first data group DA1 and Table 51 shows the second data group DA2.

  In the following description, the first data group DA1 is appropriately referred to as a non-defective product (OK product) data group, and the second data group DA2 is referred to as a defective product (NG product) data group.

[Step 3]
Next, the data string extraction unit 5 extracts each data string of the input attribute xj (1 ≦ j ≦ 4) from the good product data group DA1 (Table 50) (S3). This data string is called a 1-xj data group.

  Similarly, the data string extraction unit 5 extracts each data string of the input attribute xj (1 ≦ j ≦ 4) from the defective product data group DA2 (Table 51) (S3). This data string is called a 2-xj data group.

  The 1-xj data group is shown in Tables 52 to 55, and the 2-xj data group is shown in Tables 56 to 59.

[Step 4]
The frequency calculation unit 6 inputs each of the 1-xj data group extracted from the non-defective product data group DA1 in step 3 and each of the 2-xj data group extracted from the defective product data group DA2 in step 3 to the input attribute xj. The values are sorted in ascending order by the value of (sorting process 1). For each numerical value of the input attribute xj, 1-xj frequency cumulative% (first frequency) representing the ratio of the number of data whose input attribute xj is equal to or less than the numerical value in the non-defective product data group DA1, and defective product data In the group DA2, 2-xj frequency cumulative% (second frequency) representing the ratio of the number of data whose input attribute xj is equal to or less than the numerical value is calculated (S4).

Here, the tables 60 to 63 in which the tables 52 to 55 are rearranged in ascending order by the value of the input attribute xj are used, and the number of data at the position equal to or higher than the position of the data in the table for each row (id) data. The ratio of one data group to the total number of data (= 8) is calculated as 1-xj frequency cumulative%. Similarly, using Tables 64-68 in which Tables 56-59 are rearranged in ascending order by the value of the input attribute xj, the number of data at the position equal to or higher than the position of the data in the table for each row (id) data. The ratio of 2 data groups to the total number of data (= 4) is calculated as 2-xj frequency cumulative%. The values of 1-xj frequency cumulative% and 2-xj frequency cumulative% calculated here are shown in Tables 60 to 67. Show.

  In Steps 3 and 4 described above, after extracting and rearranging the data strings, 1-xj frequency cumulative% and 2-xj frequency cumulative% are calculated. The 1-xj frequency accumulation% and the 2-xj frequency accumulation% may be directly calculated without performing the above.

  Further, the frequency calculation unit 6 is a table of 1-xj data groups that are non-defective product data groups for which 1-xj frequency cumulative% is calculated, and 2-items that are defective product data groups for which 2-xj frequency cumulative% is calculated. The table of the xj data group is combined. Specifically, for the input attribute x1, Table 60 and Table 64 are combined to generate the x1 frequency accumulation table in Table 68, and for the input attribute x2, Table 61 and Table 65 are combined to generate the x2 frequency accumulation in Table 69. For the input attribute x3, the table 62 and the table 66 are combined to generate an x3 frequency accumulation table in Table 70, and for the input attribute x4, the table 63 and the table 67 are combined to generate an x4 frequency accumulation table in Table 71. , Create each.

  Further, the frequency calculation unit 6 rearranges the frequency accumulation tables in Tables 68 to 71 in ascending order by the value of the input attribute xj (sorting process 2). After the rearrangement process 2, the value immediately above (the value of the data on one line) is assigned to the blanks of 1-xj frequency accumulation% and 2-xj frequency accumulation% in order from the upper blank (substitution process). . Thereafter, only the data of the last line among the lines with the same value in the input attribute xj is adopted (duplicate processing). In this way, as shown in Tables 72 to 75, the frequency calculation unit 6 sets the input attribute xj to be less than or equal to the value in the first data group DA1 that is a non-defective data group for each value of the input attribute xj. 1-xj frequency cumulative percentage (A; first frequency) representing the ratio of a certain number of data and the ratio of the number of data whose input attribute xj is less than or equal to the numerical value in the second data group DA2 which is a defective product data group And 2-xj frequency cumulative percentage (B; second frequency) representing the two are calculated (S4).

[Step 5]
Next, the frequency cumulative difference calculation unit 7 calculates, for each value of the input attribute xj, a difference between 1-xj frequency cumulative% (A) of a non-defective product and 2-xj frequency cumulative% (B) of a defective product (= | A−B |) is calculated (S5). This difference value is referred to as xj frequency cumulative difference% (= | A−B |). The calculation results of the xj frequency cumulative difference% are shown in Table 72 to Table 75.

  The relationship between the value of the input attribute xj and the non-defective 1-xj frequency cumulative% (A), the defective 2-xj frequency cumulative% (B), and the xj frequency cumulative difference% | A-B | It is shown in FIG.

  The xj frequency cumulative difference% | AB | with respect to each numerical value of the input attribute xj is the first data of the non-defective product by bifurcation between the range where the input attribute xj is less than the numerical value and the range where the input attribute xj exceeds the numerical value. This is an index indicating whether the group DA1 and the defective second data group DA2 are well separated. In other words, the xj frequency cumulative difference% | A−B | is the degree to which data whose input attribute is less than or equal to the numerical value is biased to one of the first data group and the second data group, and the input attribute is This is a threshold evaluation index that comprehensively represents the degree to which the data exceeding the numerical value is biased to the other of the first data group and the second data group.

  That is, the xj frequency cumulative difference% with respect to each numerical value of the input attribute xj is “input attribute is an input attribute condition” with respect to an input attribute condition “input attribute xj is equal to or lower than the numerical value” or “input attribute xj exceeds the numerical value”. Is the data belonging to the second data group in the analysis data group, and if the input attribute does not satisfy the input attribute condition, it is data belonging to the first data group in the analysis data group. " It can be regarded as an input attribute condition evaluation index representing the probability of the rule.

  In the above, xj frequency cumulative difference% | A−B | is calculated as a threshold evaluation index. However, as a threshold evaluation index for each numerical value, an index for evaluating the degree of data bias, for example, information gain (gain ), Information gain ratio, Gini index, mean square error, or the like may be used.

[Step 6]
The input attribute threshold value determination unit 8 extracts, for each input attribute xj, the value of the input attribute xj when the value of the xj frequency cumulative difference% | AB | is the maximum among the individual values of xj. (S6). This value is called an input attribute threshold value xj-th.

  As can be seen with reference to FIGS. 3 to 6, the input attribute threshold value xj-th is divided into a range of xj ≦ xj-th and a range of xj> xj-th, so that the non-defective first data group DA1. And the value of the input attribute xj that makes it easy to separate the defective product from the second data group DA2.

  Here, the processing of the third step to the sixth step is collectively performed for a plurality of input attributes. However, the value of j is sequentially increased from 1 to N, and the processing of the third step to the sixth step is performed. May be repeated.

[Step 7]
Next, the frequency cumulative ratio calculation unit 16 calculates the ratio of the defective product 2-xj frequency cumulative% (B) to the non-defective 1-xj frequency cumulative% (A) at xj = xj-th (S7). ). This ratio is called a 2-xjth lower ratio (= B / A). Further, the value obtained by subtracting the 2-xj frequency cumulative percentage (B) of defective products from 100 (= 100-B) with respect to the value obtained by subtracting the 1-xj frequency cumulative percentage (A) of good products from 100 (= 100-A). ) Is calculated (S7). This ratio is referred to as a 2-xjth upper ratio (= (100−B) / (100−A)). Then, a 2-xjth ratio representing the larger value of both ratios is extracted.

  Here, the 2-xjth lower ratio represents a ratio at which the defective second data group can be detected separately from the first non-defective data group based on the input attribute condition “xj ≦ xj−th”. The 2-xjth upper ratio represents a ratio at which the defective second data group can be detected separately from the first non-defective data group based on the input attribute condition “xj> xj-th”.

  In other words, the 2-xjth lower ratio is an evaluation value (the first value representing the probability of the association rule that “if the input attribute xj is equal to or less than the input attribute threshold xj-th, it is data included in the second data group”). Rule evaluation value). Further, the 2-xjth upper ratio is an evaluation value (second value) indicating the probability of the association rule “if the input attribute xj exceeds the input attribute threshold value xj-th, it is data included in the second data group”. Rule evaluation value).

  The input attribute threshold value determination unit 8 determines (extracts) each input attribute xj, and the input attribute threshold value xj-th, xj = xj-th, 1-xj frequency cumulative percentage (A) of non-defective products, 2 of defective products -Xj frequency cumulative% (B), xj frequency cumulative difference% | AB |, 2-xjth lower ratio B / A, 2-xjth upper ratio (100-B) / (100-A), 2-xjth ratio Table 76 shows these values.

[Step 8]
The second factor extraction unit 9 extracts an input attribute that maximizes the 2-xjth ratio in step 7 from the input attributes x1 to x4 (S8). As a result, the input attribute that maximizes the 2-xjth ratio, the threshold value, and the type of the employed ratio (upper and lower) are data indicating the cause of the output attribute condition (input attribute condition) corresponding to the second data group. Will be extracted. This is equivalent to extracting data indicating an input attribute condition of a correlation rule having the highest 2-xjth lower ratio or 2-xjth upper ratio among the correlation rules for all input attributes.

  Here, the 2-xjth ratio is calculated as an index for extracting the input attribute of the correlation rule having the maximum rule evaluation value, but the input attribute of the correlation rule having the maximum rule evaluation value is extracted. As other indicators, other evaluation indicators such as support rate (support), certainty factor (confidence), information gain (gain), information gain ratio, Gini index, mean square error and the like may be used.

  Referring to Table 76, when input attribute x2 = x2-th = 2, 2-x2th ratio = 2-x2th upper ratio = ∞. This indicates that, under the input attribute condition “x2> 2,” the second data group DA2 of defective products can be detected completely separated from the first data group DA1 of non-defective products. 4 is easier to understand.

  The extracted data of the input attribute (= x2), the input attribute threshold value (= 2) representing the value of the input attribute, and the adopted ratio type (= top) are stored in the analysis result data storage unit 14.

As described above, the input attribute condition “x2> 2” is extracted as one factor of the problem event (the second data group DA2 of defective products).
[Step 9]
In step 8, the input attribute condition “x2> 2” is extracted as one factor of the problem event (second data group DA2 of defective products). Next, another factor is investigated. Therefore, the factor-undiscovered data extraction unit 10 includes the basic data group DA0 (Table 49) in the data group (factor data group) that satisfies the input attribute condition “x2> 2” and the basic data group DA0 (Table 49). And a data group that has not yet found the cause of the problem event (other data group), that is, a data group that satisfies the input attribute condition “x2 ≦ 2” (does not satisfy the input attribute condition “x2> 2”), A data group for which the cause of the problem event has not yet been found is extracted (S9; see Table 77).

  The factor undiscovered data extraction unit 10 sends the extracted data group to the data classification unit 4 as the next (new) basic data group DA0.

[Step 10]
Then, the data group extracted in step 9 is set as the next basic data group DA0, and the processes in steps 2 to 9 are repeated until the end condition determining unit 11 determines that the end condition is satisfied. That is, the data group extracted in step 9 is set as the next basic data group DA0, and the above-described processing in steps 2 to 9 is repeated until the end condition determination unit 11 determines that the end condition is satisfied. The end condition determination unit 11 according to the present embodiment determines that the end condition is a case where the number of data in the second data group DA2 of defective products becomes 0 in the above-described step 2 during the iterative process. As described above, detailed factor analysis results can be obtained by repeatedly performing the process until the number of data in the second data group DA2 of defective products becomes zero.

  Note that the end condition is another end condition based on the number of data in the second data group DA2, for example, (1) When the number of data in the second data group DA2 is equal to or less than a predetermined number in the above step 2 during the repetitive processing. (2) When the ratio of the number of data in the second data group DA2 to the number of data in the first data group DA1 is equal to or less than a predetermined ratio in the above step 2 during the iterative process, (3) step 8 during the iterative process. The rule evaluation value of the input attribute condition extracted in step 1 may be lower than a predetermined value. When such termination conditions are used, a simpler and sufficient factor analysis result can be obtained. Further, when priority is given to obtaining a concise factor analysis result, the end condition is simply a case where the iterative process is performed a predetermined number of times, or the end condition determining unit 11 is omitted and the iterative process is performed as much as possible. May be.

  In this example, defective data (second data group DA2; y = 2) is not included in the data group of x1 ≦ 2 that has not been found and extracted in step 9 during the second iteration. Therefore, the iterative process was completed at the second time (after the factor extraction was performed for the second time).

[Step 11]
The input attribute threshold value table creation unit 12 creates an input attribute threshold value table that stores the input attribute xj extracted for each repetition process of step 10, the input attribute threshold value xj-th, and the type of ratio adopted ( S11; see Table 78).

The input attribute threshold value table creating unit 12 converts the numerical value of the input attribute threshold value xj-th in the input attribute threshold value table into character data as necessary. The conversion rule for character data is a rule that is the reverse conversion of the conversion in step 0, and is as follows.
(X1) 1 → A, 2 → B, 3 → C, 4 → D
(X2) 1 → a, 2 → b, 3 → c, 4 → d
(X3) Not converted (x4) Not converted Table 79 shows an input attribute threshold value table in which the input attribute threshold value xj-th in the input attribute threshold value table of Table 78 is converted into character data.

  This input attribute threshold value table corresponds to the classification conditions of the decision tree when focusing on the separation of the output attribute y = Y (y = 2) in the conventional decision tree-2 described in Patent Document 1 (FIG. 12). To do.

[Step 12]
Next, the contribution rate calculating unit 13 extracts the input attribute condition extracted from the input attribute threshold value table of Table 78 for the problem event (y = 2: the original second data group DA2 which is a defective product data group). The contribution ratio (the ratio of the number of defects due to the input attribute condition in the total number of defects) is obtained.

  Table 80 shows the input attribute condition “x2> 2” extracted as the first factor as the factor in the original second data group DA2 (Table 51), which is a problem event (defective product), or extracted the second time. The data corresponding to the input attribute condition “x1> 2” is added with “*”.

  From Table 80, the contribution ratios of the input attribute conditions “x1> 2” and “x2> 2” for the problem event (original second data group DA2) are obtained as shown in Table 81.

  In Table 81, the contribution ratio shown at the intersection of “x1> 2” and “x1> 2” and the contribution ratio shown at the intersection of “x2> 2” and “x2> 2” are “x1>”, respectively. 2 ”represents the contribution rate of the single factor, and“ x2> 2 ”represents the contribution rate of the single factor. In addition, the contribution ratios shown at the intersections of “x1> 2” and “x2> 2” all represent the contribution ratios of the combined factors of the “x1> 2” factor and the “x2> 2” factor. The table 81 can also be expressed as shown in FIG.

[Step 13]
The data analysis is thus completed, and the input attribute threshold value table created by the input attribute threshold value table creating unit 12 and the contribution rate data are stored as analysis result data in the analysis result data storage unit 14 such as a hard disk. The analysis result data is appropriately sent from the analysis result data storage unit 14 to the output unit 15 such as a display device or a printing device, and is displayed on the display device as a decision tree or table, or is determined by the printing device. Can be printed as.

  According to the present embodiment, as in the conventional decision tree-2 (FIG. 12) described in Patent Document 1, the label hierarchy structure (FIG. 11) is not defined in advance, but the table 79 (or table 78) can be obtained. The cause of the problem event can be derived in a very simple form as shown in the input attribute threshold value table. Then, using this, the contribution rate of each factor (input attribute) to the problem phenomenon can be obtained.

  Here, when the input attribute threshold value table of this embodiment shown in Table 79 (or Table 78) is expressed in the form of a decision tree, it is expressed as shown in FIG. Also, when the contribution rate of each factor to the problem event y = Y (= 2) is expressed in the same format as FIG. 7 using the conventional decision tree-2 (FIG. 12), it is as shown in FIG.

  When the decision tree derived from the present embodiment (FIG. 8) is compared with the conventional decision tree-2 (FIG. 12), the contribution of the input attribute x3 is not expressed in the present embodiment. This is because the problem event y = Y (y = 2) does not occur due to the single factors of the input attributes x1 and x3, as can be seen by comparing FIG. 7 and FIG. This is because, in step 9 during the second repetitive operation, the repetitive process (step 10) is not executed for the data group of x1> 2.

  When pursuing factors in detail, it is also necessary to extract the contribution of the input attribute x3. However, for the purpose of removing (improving) the problem event y = Y (y = 2), only the input attribute x1 is required. Even this extraction can sufficiently achieve this purpose. In the present embodiment, paying attention to this point, the main factor that should be taken against the problem phenomenon is extracted, so the input attribute x3 is not extracted. If a detailed analysis is required, iterative processing (step 10) may be performed on both of the data groups divided in step 9 above.

  In the above-described embodiment, a plurality of factors are derived and a decision tree is generated. However, if only one factor is desired to be extracted, the process may end in step 8.

  The data analysis method described above can be realized by the computer executing a data analysis program including processes corresponding to S0 to S12 (steps 0 to 13) in FIG. Therefore, in the data analysis apparatus of FIG. 1, the data analysis program includes a computer that converts a character-numeric data conversion unit 1, an analysis target data storage unit 2, a threshold setting unit 3, a data classification unit 4, a data string extraction unit 5, a frequency calculation. Unit 6, frequency cumulative difference calculation unit 7, input attribute threshold value determination unit 8, frequency cumulative ratio calculation unit 16, second factor extraction unit 9, factor undiscovered data extraction unit 10, end condition determination unit 11, input attribute threshold value table This can be realized by functioning as the creation unit 12 and the contribution rate calculation unit 13.

  The program can be provided to the user by storing it in a computer-readable recording medium. The recording medium may be a built-in medium built in the computer main body, or a removable medium configured to be separable from the computer main body. Examples of the built-in medium include ROM; rewritable nonvolatile memory such as flash memory; and hard disk. In addition, as the removable media, optical recording media such as CD-ROM and DVD; magneto-optical recording media such as MO; magnetic recording media such as floppy (registered trademark) disk, cassette tape, and removable hard disk; memory cards and the like And a medium having a built-in rewritable nonvolatile memory such as a medium having a built-in ROM such as a ROM cassette.

  The program may be configured to be executed by CPU access, or after reading the program stored in the recording medium and transferring the read program to the program storage area of the built-in medium, the program on the built-in medium May be executed by CPU access. In addition, the program is not limited to be sold in a state where it is stored in a computer-readable recording medium, and is sold in a format that is transferred to a user's computer via a communication network such as the Internet. It may be a thing.

  In this embodiment, the data classification unit 4 classifies the output attribute by comparing the output attribute with the output attribute threshold value. However, when the output attribute is a character attribute, the character-numeric data conversion unit 1 sets the output attribute to a numerical value. Instead of converting into attributes, the data classification unit 4 may perform classification by comparing the output attributes with the output attributes (characters; Y) to be analyzed.

As described above, the data analysis method according to the present embodiment includes N columns of input attribute data including N attributes (N is an integer of 2 or more) and one column of output attributes including one attribute. A data analysis method for analyzing a causal relationship between the output attribute and the input attribute, and a first step of setting an output attribute threshold; A second step of dividing the basic data group into a first data group and a second data group based on a comparison between the value and the output attribute threshold value; and the first data group and the second data group A third step of extracting a 1-J data string and a 2-J data string each representing a data string of a Jth input attribute (J is an integer having a relationship of 1 ≦ J ≦ N); -For each value of the Jth input attribute of the J data string, 1-J frequency accumulation (%) representing the ratio of the number of data of 2 is calculated, and for each value of the Jth input attribute of the 2-J data string, 2 representing the ratio of the number of data less than that value -For the fourth step of calculating the J frequency accumulation (%) and for each individual value of all of the Jth input attributes, including both the 1-J data string and the 2-J data string, The fifth step of calculating the J-th frequency cumulative difference, which represents the absolute value of the difference between the 1-J frequency cumulative (%) and the 2-J frequency cumulative (%), and the value of the J-th frequency cumulative difference is the maximum The sixth step of extracting the value of the J-th input attribute as the J-th input attribute threshold value, and the second step with respect to the 1-J frequency accumulation (%) when the J-th input attribute is the J-th input attribute threshold value -J lower ratio representing the ratio of J frequency accumulation (%), and subtracting 1-J frequency accumulation (%) from 100 The 2-J upper ratio representing the ratio of the value obtained by subtracting the 2-J frequency accumulation (%) from 100 is calculated, and the 2-J ratio indicating the larger value of both ratios is extracted. The seventh step and the value of J are sequentially increased from 1 to N, the operations of the third step to the seventh step are repeated, and the first to Nth extracted in the seventh step during the repeated operation An eighth step of extracting and storing the input attribute having the maximum value, the input attribute threshold representing the value of the input attribute, and the type of the adopted ratio among the 2-J ratio of the input attribute up to Based on the input attribute extracted in the eighth step, at least one of the ninth step for dividing the basic data group into two and the data group divided in the ninth step is used as a new basic data group. Until the predetermined termination condition is satisfied. And a tenth step for repeating the operation of the ninth step.

  According to the above method, it is possible to derive a plurality of factors of problem events in a very simple form without defining the label hierarchical structure in advance. Then, by using this, it is possible to create a decision tree representing a causal relationship, and obtain the contribution rate of each factor (input attribute) to the problem event (output attribute).

  The present invention provides an output attribute (object attribute) to be analyzed, such as characteristics of a product manufactured in a manufacturing process, and an input attribute (explanatory attribute) that is an attribute affecting the output attribute, such as a manufacturing process condition. Can be used to determine the input attribute condition for the output attribute value to be collected, or to analyze the causal relationship between the input attribute and the output attribute. Therefore, this invention can be utilized for improvement of the manufacturing process in the manufacturing industry, for example.

It is a block diagram which shows the structure of the data analyzer which concerns on one Embodiment of this invention. It is a flowchart which shows the data analysis method which concerns on one Embodiment of this invention. FIG. 7 is a graph showing an example of the output of the frequency accumulation difference calculation unit 7 in the data analysis apparatus according to the embodiment of the present invention, and includes an input attribute x1, a non-defective 1-x1 frequency accumulation (A), and a non-defective 2- The relationship between x1 frequency accumulation (B) and x1 frequency accumulation difference | AB | is shown. FIG. 7 is a graph showing an example of the output of the frequency cumulative difference calculation unit 7 in the data analysis apparatus according to the embodiment of the present invention. The input attribute x2, the non-defective 1-x2 frequency accumulation (A), and the defective 2- The relationship between x2 frequency accumulation (B) and x2 frequency accumulation difference | AB | is shown. FIG. 7 is a graph showing an example of the output of the frequency cumulative difference calculation unit 7 in the data analysis apparatus according to the embodiment of the present invention. The input attribute x3, the non-defective 1-x3 frequency accumulation (A), and the defective 2- The relationship between x3 frequency accumulation (B) and x3 frequency accumulation difference | AB | is shown. FIG. 6 is a graph showing an example of the output of the frequency accumulation difference calculation unit 7 in the data analysis apparatus according to the embodiment of the present invention, where an input attribute x4, a non-defective 1-x4 frequency accumulation (A), and a non-defective 2- The relationship between x4 frequency accumulation (B) and x4 frequency accumulation difference | AB | is shown. It is an example of the data output by the contribution rate calculating part 13 (step 12) in the data analyzer which concerns on one Embodiment of this invention, and the input attribute condition "with respect to the output attribute condition y = 2 (= Y) which is a problem event" x1> 2 ”and the contribution ratio of the input attribute condition“ x2> 2 ”. It is the figure which expressed the input attribute threshold value table of the embodiment of the present invention in the form of a decision tree. It is the figure which expressed the conventional decision tree-2 in the same format as FIG. It is a figure showing the conventional decision tree-1. It is a figure showing the label hierarchical structure of the conventional decision tree-2, (a) shows x1 attribute, (b) shows x2 attribute, (c) shows x3 attribute, (d) shows x4 attribute. It is a figure showing the conventional decision tree-2. It is a block diagram which shows the structure of the input attribute condition determination apparatus and data analysis apparatus which concern on other embodiment of this invention. It is a flowchart which shows the data analysis method which concerns on other embodiment of this invention. It is a flowchart which shows the input attribute condition determination method which concerns on other embodiment of this invention. It is the figure which expressed an example (Table 35) of the data output by the inferior goods isolation | separation degree calculating part 112 (step 109) in the data analyzer which concerns on other embodiment of this invention with the Venn diagram. It is the figure which expressed the process of the factor extraction (step 110) and the factor determination (step 114) of the data analysis method which concerns on other embodiment of this invention in the form of the decision tree. Regarding the example of the determination factor list table (Table 47) output by the factor determination unit 117 (step 114) in the data analysis apparatus according to another embodiment of the present invention, the number of defects for each input attribute condition is represented by a bar graph. It is the figure which expressed separation degree (2nd data group separation degree) with the line graph. Using an example (Table 46) of data output from the composite factor defect number calculation unit 118 (step 115) in the data analysis apparatus according to another embodiment of the present invention, the number of defects for each input attribute condition is represented by a bar graph. A non-defective product separation degree (second data group separation degree) is represented by a line graph, and hatching is performed on the number of defects due to a composite factor with the first factor (“x2> 2”, ie, “x2 = c or d”). It shows with. It is a graph explaining the Gini index method which is the prior art described in Non-Patent Document 1, and shows the relationship between the branch condition of the input attribute x1 and the improvement degree of the Gini index method using the data group of Table 1 as the subject. is there. It is a graph explaining the Gini index method which is the prior art described in Non-Patent Document 1, and shows the relationship between the branch condition of the input attribute x2 and the improvement degree of the Gini index method using the data group of Table 1 as the subject. is there. It is a graph explaining the Gini index method which is the prior art described in Non-Patent Document 1, and shows the relationship between the branch condition of the input attribute x3 and the improvement degree of the Gini index method using the data group of Table 1 as the subject. is there. It is a graph explaining the Gini index method which is the prior art described in Non-Patent Document 1, and shows the relationship between the branch condition of the input attribute x3 and the improvement degree of the Gini index method using the data group of Table 1 as a subject. is there.

Explanation of symbols

1 Character-numeric data converter (numeric converter)
3 threshold setting unit (threshold setting means)
4 Data classification part (classification means)
5 Data string extraction unit 6 Frequency calculation unit (Frequency calculation means in the first evaluation means)
7 Frequency cumulative difference calculation unit (difference calculation means in the first evaluation means)
8 Input attribute threshold value determination unit (threshold value determination means)
9 Second factor extraction unit (second factor extraction means)
10 Factor undiscovered data extraction unit (division means)
11 End condition determination unit (end condition determination means)
14 analysis result data storage unit 15 output unit 16 frequency cumulative ratio calculation unit (second evaluation means, division rule evaluation means)
102 basic data group storage unit 103 classification condition setting unit (classification condition setting means)
104 Data classification part (classification means)
105 Classification basic data group storage unit 106 Analysis data group extraction unit (analysis data group extraction means)
107 data row separation unit 109 first factor extraction unit (first factor extraction means)
111 Input attribute condition determining unit (input attribute condition determining means)
112 Defective product separation degree calculation unit (second data group separation degree calculation means)
115 Data division unit (division means)
117 Factor determination unit (factor determination means)
118 Compound Factor Failure Number Calculation Unit 119 Numerical Value-Character Data Conversion Unit 130 Threshold Determination Unit (Threshold Determination Unit)
131 Polarity determination unit (polarity determination means)

Claims (27)

A set of data composed of at least one input attribute that is a numerical attribute and an output attribute, and an analysis data group that is classified into a first data group and a second data group according to the value of the output attribute The analysis data group is bifurcated so that the first data group is biased to one data group and the second data group is biased to the other data group among the data groups obtained by the bisection based on the threshold value of the input attribute. An input attribute condition determination device for determining an input attribute condition, which is an input attribute condition for
Every few values the input attributes to obtain Ri bets, among the data having the following values the numerical, the number of data belonging to the first data group, which is the ratio to the number of all data belonging to the first data group first For every numerical value that can be taken by the input attribute, the number of data belonging to the second data group out of all the data belonging to the second data group is calculated. Frequency calculating means for calculating a second frequency that is a ratio to the number of data of
Difference calculation means for calculating a difference value between the first frequency and the second frequency for each of all numerical values taken by the input attribute;
Threshold value determining means for determining a numerical value that maximizes the difference value among the numerical values taken by one input attribute as a threshold value in the input attribute, and determining at least one threshold value corresponding to at least one input attribute;
An input attribute condition determining device, comprising: an input attribute condition determining unit that determines the input attribute condition based on the threshold value determined by the threshold value determining unit. A polarity determining means for determining a magnitude relationship between the first frequency and the second frequency in the threshold determined by the threshold determining means;
The input attribute condition determining means is
The second data group is biased toward the data group that satisfies the input attribute condition, and the first data group is biased toward the data group that does not satisfy the input attribute condition.
If the polarity determining means determines that the first frequency is greater than the second frequency, the input attribute condition is determined as a condition that “the input attribute exceeds the threshold”,
2. The input attribute condition is determined to be a condition that “the input attribute is equal to or less than a threshold value” when the polarity determination unit determines that the second frequency is greater than the first frequency. The described input attribute condition determination device. The input attribute is a manufacturing process condition and / or an in-line inspection result in the product manufacturing process,
The above output attribute is the product quality judgment result,
The input attribute condition determining apparatus according to claim 1, wherein the second data group is a data group having a bad quality determination result. A plurality of input attributes, the basic data group is a set of data composed by the output attribute, a causal relationship between the input attributes and output attributes A data analyzer you analysis,
Classifying means for classifying the basic data group into a first data group and a second data group according to the value of the output attribute, and assigning a classification flag;
An analysis data group extraction means for extracting an analysis data group to be analyzed from the basic data group after the classification;
An input attribute condition determining device according to claim 1 or 2,
The frequency calculation means and the difference calculation means perform the calculation for each of all the numerical values taken by each input attribute of the analysis data group,
The threshold value determining means determines a threshold value for each input attribute of the analysis data group, respectively.   5. The data analysis apparatus according to claim 4, further comprising numerical conversion means for performing processing for converting an input attribute into a numerical value for a basic data group including an input attribute that is not a numerical attribute. Polarity determination means for determining a magnitude relationship between the first frequency and the second frequency in the threshold value determined by the threshold value determination means;
For each of the input attribute conditions determined by the input attribute condition determination device, if the polarity determination means determines that the second frequency is greater than the first frequency, the second frequency relative to the first frequency When the first frequency is determined to be greater than the second frequency by the polarity determination means, the (100% -first frequency) ( Evaluation value calculating means for calculating a second ratio, which is a ratio of 100% −second frequency), as an evaluation value ;
Among the input attribute conditions determined by the input attribute condition determination device, based on the input attribute condition having the maximum evaluation value , the analysis data group, the factor data group satisfying the input attribute condition, and the input The data analysis apparatus according to claim 4, further comprising a dividing unit that divides the data into other data groups that do not satisfy the attribute condition. The analysis data group extraction means extracts at least one of the data groups divided by the division means as a new analysis data group,
A series of processes consisting of a process by an analysis data group extracting unit, a process by an input attribute condition determining device, a process by an evaluation value calculating unit, and a process by a dividing unit are repeatedly executed. Item 7. The data analysis device according to Item 6.   8. The data analysis apparatus according to claim 7, wherein the analysis data group extraction unit extracts only another data group from the data group divided by the division unit as a new analysis data group. . It further includes a classification condition setting means for setting a classification condition,
5. The data analysis apparatus according to claim 4, wherein the classification unit classifies the basic data group based on the classification condition set by the classification condition setting unit. The basic data group includes a plurality of output attributes,
The classification condition setting means sets a classification condition for each of the plurality of output attributes,
10. The data according to claim 9 , wherein the classification means classifies the basic data group based on a logical sum or logical product of the classification conditions set by the classification condition setting means. Analysis equipment. For each of the input attribute conditions determined by the input attribute condition determination device, a second data group separation degree representing a ratio of the second data group included in the data satisfying the input attribute condition in the basic data group Second data group separation degree computing means for computing
Among the input attribute conditions determined by the input attribute condition determination device, the second data group separation degree having a value larger than the second data group content ratio representing the ratio of the second data group in the basic data group is set. data analysis device according to any one of claims 4-7, characterized in that it comprises a first factor extraction means to extract an input attribute condition with. In the input attribute conditions determined by said input attribute condition determining apparatus, in claim 6 or 7, characterized in that it comprises a second factor extraction means to extract an input attribute condition with the largest evaluation value The data analysis device described. A plurality of input attributes, the basic data group is a set of data composed by the output attribute, a causal relationship between the input attributes and output attributes A data analyzer you analysis,
Classifying means for classifying the basic data group into a first data group and a second data group according to the value of the output attribute, and assigning a classification flag;
An analysis data group extraction means for extracting an analysis data group to be analyzed from the basic data group after the classification;
For each input attribute condition that can be taken by each input attribute of the analysis data group, “if the input attribute satisfies the input attribute condition, the data belongs to the second data group in the analysis data group and the input attribute is input. A first evaluation unit that calculates an input attribute condition evaluation index that represents the probability of the first association rule that if the attribute condition is not satisfied, the data belongs to the first data group in the analysis data group;
For each input attribute of the analysis data group, an input attribute condition determining means for determining an input attribute condition having the maximum input attribute condition evaluation index as an input attribute condition satisfying the first correlation rule;
For each of the input attribute conditions determined by the input attribute condition determining means, a second correlation rule that “if the input attribute satisfies the input attribute condition, it is data included in the second data group in the analysis data group” A second evaluation means for calculating a second evaluation index representing the certainty of
In the input attribute conditions determined by said input attribute condition determining means, viewed contains a factor extraction unit second evaluation index to extract an input attribute condition having the maximum
The first evaluation means includes
For each value that each input attribute can take, a first value that is a ratio of the number of data belonging to the first data group to the number of all data belonging to the first data group among data having a numerical value equal to or less than the value All the data belonging to the second data group, the number of data belonging to the second data group among the data having a numerical value equal to or lower than the numerical value for each numerical value that can be calculated and the input attribute can be obtained Frequency calculating means for calculating a second frequency that is a ratio to the number of
A data analysis apparatus comprising difference calculation means for calculating a difference between a first frequency and a second frequency for all numerical values of each input attribute . A plurality of input attributes, and analyzed the basic data group is a set of data composed by the output attribute, a causal relationship between the input attributes and output attributes A data analyzer you analysis,
A classifying means for classifying the basic data group into a first data group and a second data group according to output attributes;
First evaluation means for calculating a threshold evaluation index representing the degree to which data having a numerical value less than or equal to the numerical value is biased to one of the first data group and the second data group for each numerical value of each input attribute When,
Threshold determination means for determining a numerical value having the maximum threshold evaluation index for each input attribute as a threshold of each input attribute based on the threshold evaluation index calculated by the first evaluation means;
Based on the threshold value of each input attribute determined by the threshold value determining means, the first rule evaluation value representing the probability of the association rule that “the input attribute is equal to or less than the threshold value is data included in the second data group” And a second rule evaluation value that calculates the second rule evaluation value representing the probability of the association rule that “if the input attribute exceeds the threshold, the data is included in the second data group” for each input attribute. When,
Rule evaluation value by the second evaluation means is computed, including a factor extraction unit to extract the data representing the input attribute condition of the correlation rule with the highest rule evaluation value among the correlation rules all input attributes See
The first evaluation means includes
For each value that each input attribute can take, a first value that is a ratio of the number of data belonging to the first data group to the number of all data belonging to the first data group among data having a numerical value equal to or less than the value All the data belonging to the second data group, the number of data belonging to the second data group among the data having a numerical value equal to or lower than the numerical value for each numerical value that can be calculated and the input attribute can be obtained Frequency calculating means for calculating a second frequency that is a ratio to the number of
A data analysis apparatus comprising difference calculation means for calculating a difference between a first frequency and a second frequency for all numerical values of each input attribute . Based on the input attribute condition extracted by the factor extraction means , further includes a dividing means for dividing the analysis data group into a factor data group satisfying the input attribute condition and another data group not satisfying the input attribute condition. ,
The analysis data group extraction means extracts at least one of the data groups divided by the division means as a new analysis data group,
A series of processes consisting of processing by the analysis data group extracting means, processing by the first evaluating means, processing by the input attribute condition determining means, processing by the second evaluating means, processing by the factor extracting means , and processing by the dividing means is repeated. 14. The data analysis apparatus according to claim 13 , wherein the data analysis apparatus is executed. The data analysis apparatus according to claim 15 , wherein the analysis data group extraction unit extracts only another data group from the data group divided by the division unit as a new analysis data group. . The input attribute is a manufacturing process condition and / or an in-line inspection result in the product manufacturing process,
The above output attribute is the product quality judgment result,
It said second data group, the data analyzer according to claim 13 or 14, characterized in that the quality determination result is data group bad. It further includes an end condition determining means for determining whether or not the end condition is satisfied,
The end condition determining means determines whether the number of data of the second data group in the analysis data group extracted by the analysis data group extracting means is 0 as an end condition,
16. The data analysis apparatus according to claim 7, wherein when the end condition determining unit determines that the end condition is satisfied, the execution of the series of processes is ended. A set of data composed of at least one input attribute that is a numerical attribute and an output attribute using the input attribute condition determination device according to claim 1, and the first data group depending on the value of the output attribute With respect to the analysis data group classified as the second data group, the first data group is biased to one data group among the data groups obtained by the bisection based on the threshold value of the input attribute, and the first data group is the second data group. An input attribute condition determination method for determining an input attribute condition, which is an input attribute condition for bisecting the analysis data group so that two data groups are biased ,
By the frequency calculating means, every several values to obtain Ri said input attributes bets, among the data having the following values the numerical, the number of data belonging to the first data group, of all the data belonging to the first data group The number of data belonging to the second data group among the data having a numerical value equal to or lower than the numerical value for each numerical value that can be taken by the input attribute is calculated. A frequency calculating step for calculating a second frequency that is a ratio to the number of all data belonging to the two data groups ;
A difference calculating step for calculating a difference value between the first frequency and the second frequency for each of all the numerical values taken by the input attribute by the difference calculating means;
The threshold value determining means determines a numerical value that maximizes the difference value among the numerical values taken by one input attribute as a threshold value for the input attribute, and determines at least one threshold value corresponding to at least one input attribute. A threshold determination step;
An input attribute condition determination method comprising: an input attribute condition determination step for determining the input attribute condition based on the threshold value determined by the threshold value determination means by the input attribute condition determination means. Using the data analysis device according to claim 4, a plurality of input attributes and, to the basic data group is a set of data composed by the output attribute, analyze the causal relationship between input attributes and output attributes a to that data analysis methods,
A step of classifying the basic data group into a first data group and a second data group according to an output attribute value by the classification means, and assigning a classification flag;
An analysis data group extraction step of extracting the analysis data group to be analyzed from the basic data group after the classification by the analysis data group extraction means;
By the frequency calculating means of the input attribute condition determining apparatus, every few values each input attribute analysis data group obtained Ri bets, among the data having the following values the numerical number of data belonging to the first data group Among the data having a numerical value equal to or lower than the numerical value for each numerical value that can be taken by the input attribute, and calculating the first frequency that is a ratio to the number of all data belonging to the first data group A frequency calculating step for calculating a second frequency that is a ratio of the number of data belonging to the second data group to the number of all data belonging to the second data group ;
A difference calculating step of calculating a difference value between the first frequency and the second frequency for each of all the numerical values taken by the input attribute by the difference calculating means of the input attribute condition determining device;
A threshold value determining step of determining, for each input attribute, a numerical value that maximizes the difference value as a threshold value of the input attribute by the threshold value determining means of the input attribute condition determining device;
The input attribute condition determining means of the input attribute condition determining device sets a first data group out of data groups obtained by bisection based on the threshold value of the input attribute based on the threshold value determined by the threshold value determining means . And an input attribute condition determining step for determining an input attribute condition for bisecting the analysis data group so that one data group is biased and the second data group is biased to the other data group. Method. A polarity determination step of determining a magnitude relationship between the first frequency and the second frequency in the threshold value determined by the threshold value determination means by the polarity determination means according to claim 6;
The evaluation value calculation means according to claim 6 determines that the second frequency is greater than the first frequency by the polarity determination means for each of the input attribute conditions determined by the input attribute condition determination device . In this case, when the first ratio, which is the ratio of the second frequency to the first frequency, is calculated as an evaluation value, and the first frequency is determined to be greater than the second frequency by the polarity determination unit. Is an evaluation value calculation step for calculating, as an evaluation value , a second ratio that is a ratio of (100% −second frequency) to (100% −first frequency) ;
By the dividing means of said data analyzer according to claim 6, in the input attribute conditions determined by said input attribute condition determining apparatus, based on the input attribute conditions with the maximum of the evaluation value, the analytical data 21. The data analysis method according to claim 20 , further comprising a dividing step of dividing the group into a factor data group that satisfies the input attribute condition and another data group that does not satisfy the input attribute condition. Using the data analysis device according to claim 13, a plurality of input attributes and, to the basic data group is a set of data composed by the output attribute, analyze the causal relationship between input attributes and output attributes a to that data analysis methods,
A step of classifying the basic data group into a first data group and a second data group according to an output attribute value by the classification means, and assigning a classification flag;
An analysis data group extraction step of extracting the analysis data group to be analyzed from the basic data group after the classification by the analysis data group extraction means;
For each of all the input attribute conditions that can be taken by each input attribute by the first evaluation means, “if the input attribute satisfies the input attribute condition, the data belongs to the second data group in the analysis data group, A first evaluation that calculates an input attribute condition evaluation index that represents the probability of the first association rule that the input attribute does not satisfy the input attribute condition is data belonging to the first data group in the analysis data group. Steps,
An input attribute condition determining step for determining, for each input attribute, an input attribute condition having the maximum input attribute condition evaluation index as an input attribute condition satisfying the first correlation rule by the input attribute condition determining means; ,
For each of the input attribute conditions determined by the input attribute condition determination means by the second evaluation means, “if the input attribute satisfies the input attribute condition, the data included in the second data group in the analysis data group A second evaluation step of calculating a second evaluation index representing the certainty of the second association rule “is”;
The factor extraction unit, in the input attribute conditions determined by said input attribute condition determining means, viewed contains a factor extraction step in which the second evaluation index to extract an input attribute condition having the maximum
The first evaluation step includes
For each numerical value that can be taken by each input attribute by the frequency calculating means, the number of data belonging to the first data group out of the data having numerical values less than or equal to the numerical value is relative to the number of all data belonging to the first data group. The second data of the number of data belonging to the second data group among the data having a numerical value equal to or lower than the numerical value for each numerical value that can be taken by the input attribute is calculated for calculating the first frequency as the ratio. A frequency calculating step for calculating a second frequency that is a ratio to the number of all data belonging to the group;
A data analysis method comprising: a difference calculation step of calculating a difference between the first frequency and the second frequency for all numerical values of each input attribute by the difference calculation means . A set of data composed of at least one input attribute that is a numerical attribute and an output attribute, and an analysis data group that is classified into a first data group and a second data group according to the value of the output attribute ,
Computer
Every few values the input attributes to obtain Ri bets, among the data having the following values the numerical, the number of data belonging to the first data group, which is the ratio to the number of all data belonging to the first data group first For every numerical value that can be taken by the input attribute, the number of data belonging to the second data group out of all the data belonging to the second data group is calculated. Frequency calculating means for calculating a second frequency that is a ratio to the number of data of
Difference calculation means for calculating a difference value between the first frequency and the second frequency for each of all the numerical values taken by the input attribute,
Threshold value determining means for determining a numerical value that maximizes the difference value among the numerical values taken by one input attribute as a threshold value in the input attribute, and determining at least one threshold value corresponding to at least one input attribute; and
Based on the threshold value determined by the threshold value determination means, the first data group is biased to one data group among the data groups obtained by the bisection based on the threshold value of the input attribute, and the second data group is the other data group. An input attribute condition determining program for functioning as an input attribute condition determining means for determining an input attribute condition, which is an input attribute condition for dividing the analysis data group into two so as to be biased . A computer-readable recording medium on which the input attribute condition determination program according to claim 23 is recorded. For a basic data group that is a set of data consisting of multiple input attributes and output attributes,
Classifying means for classifying the basic data group into a first data group and a second data group according to the value of the output attribute, and assigning a classification flag;
Analysis data group extraction means for extracting an analysis data group to be analyzed from the basic data group after the classification,
Every few values each input attribute analysis data group obtained Ri bets, among the data having the following values the numerical, the number of data belonging to the first data group, the number of all data belonging to the first data group For each value that can be taken by the input attribute, a second frequency of the number of data belonging to the second data group is calculated for each value that can be taken by the input attribute. A frequency calculating means for calculating a second frequency which is a ratio to the number of all data belonging to the data group ;
Difference calculation means for calculating a difference value between the first frequency and the second frequency for each numerical value of each input attribute of the analysis data group, and for each input attribute, the difference value is maximum. A threshold value determining means for determining a numerical value to be a threshold value of the input attribute;
Based on the threshold value determined by the threshold value determination means, the first data group is biased to one data group among the data groups obtained by the bisection based on the threshold value of the input attribute, and the second data group is the other data group. A data analysis program for functioning as an input attribute condition determining means for determining an input attribute condition, which is an input attribute condition for dividing the analysis data group into two so as to be biased . For a basic data group that is a set of data consisting of multiple input attributes and output attributes,
Computer
Classifying means for classifying the basic data group into a first data group and a second data group according to an output attribute value, and adding a classification flag;
Analysis data group extraction means for extracting an analysis data group to be analyzed from the basic data group after the classification,
For each of all the input attribute conditions that each input attribute can take, “If the input attribute satisfies the input attribute condition, the data belongs to the second data group in the analysis data group, and the input attribute satisfies the input attribute condition. 1st evaluation means for calculating an input attribute condition evaluation index, which represents the certainty of the first association rule “if there is no data belonging to the first data group in the analysis data group,”
For each input attribute, input attribute condition determining means for determining an input attribute condition having the maximum input attribute condition evaluation index as an input attribute condition satisfying the first correlation rule,
For each of the input attribute conditions determined by the input attribute condition determining means, a second correlation rule that “if the input attribute satisfies the input attribute condition, it is data included in the second data group in the analysis data group” A second evaluation means for calculating a second evaluation index representing the certainty of
In the input attribute conditions determined by said input attribute condition determining means, a data analysis program for functioning as a factor extraction means second evaluation index to extract an input attribute condition having the maximum
For each value that each input attribute can take, a first value that is a ratio of the number of data belonging to the first data group to the number of all data belonging to the first data group among data having a numerical value equal to or less than the value All the data belonging to the second data group, the number of data belonging to the second data group among the data having a numerical value equal to or lower than the numerical value for each numerical value that can be calculated and the input attribute can be obtained Frequency calculating means for calculating a second frequency that is a ratio to the number of
A data analysis program including difference calculation means for calculating a difference between a first frequency and a second frequency for all numerical values of each input attribute. 27. A computer-readable recording medium recording the data analysis program according to claim 25 and / or the data analysis program according to claim 26 .

Images