JPWO2017221856A1 - Analysis device, analysis method, and storage medium - Google Patents

Abstract

It is possible to perform data analysis with reduced influence of sample number deviation without reducing or adding sample data. The analysis device performs machine learning analysis for deriving a relationship between the explanatory variable and the objective variable of the plurality of groups for a plurality of groups generated by grouping a plurality of analysis targets in which the explanatory variable and the objective variable are associated with each other. , Based on the analysis means for each grouping, the values of the explanatory variables of the plurality of groups, and the relationship, the calculation of the predicted value that is the value of the target variable of the plurality of groups is performed for each grouping. And a calculation means for calculating a score relating to the analysis target by calculation based on the predicted value of the group to which the analysis target belongs, calculated for each grouping.

Description

  The present disclosure relates to data analysis and prediction.

  In data prediction based on machine learning using teacher data, if the number of samples differs greatly for each value of the objective variable in the teacher data, there is a problem that the accuracy of the prediction decreases.

  For example, when machine learning is performed using the number of occurrences of an event per day (for example, an accident) in a certain area as an objective variable, in general, a set of past occurrences per day and explanatory variables is a teacher. Used as data. At this time, in the majority of the teacher data, the value of the objective variable is “0” or “1”, and there are few teacher data having the value of the objective variable of “2” or more. sell. That is, in the teacher data as described above, the number of samples greatly differs for each value of the objective variable. In this way, teacher data in which the number of samples for each value of the objective variable is biased may be referred to as imbalanced data (Imbalanced Data). When machine learning and prediction are performed based on unbalanced data, the influence of sample data having a relatively low frequency of appearance of objective variable values is reduced, and the accuracy of prediction is deteriorated.

  In particular, in the above-described prediction of the number of occurrences, although it is desired to specify an area where the number of occurrences is large, samples having a large number of occurrences have a relatively small number. It is easy to be ignored in learning. As a result, for example, the formula for predicting the number of occurrences does not reflect the characteristics of the region where the number of occurrences is large, and there is a possibility that accurate prediction cannot be performed.

  As an example of a technique for solving the above problem, there is a method called Random Over Sampling (ROS) or Random Under Sampling (RUS) described in Patent Document 1.

  ROS is a method of increasing the number of samples of the class with the lower frequency in accordance with the number of samples of the class with the higher appearance frequency of the value of the objective variable among the two classes included in the teacher data. RUS is a method of reducing the number of samples of the higher frequency class in accordance with the number of samples of the lower frequency class of the two classes included in the teacher data.

JP 2010-204966 A

  ROS may generate meaningless noise data. On the other hand, RUS may exclude useful sample data. The reason is to artificially reduce or add sample data.

  An object of the present invention is to provide an analysis apparatus and method capable of performing data analysis with reduced influence of the deviation in the number of samples without reducing or adding sample data.

  An analysis apparatus according to an aspect of the present invention relates to a plurality of groups generated by grouping a plurality of analysis targets in which an explanatory variable and an objective variable are associated, and a relationship between the explanatory variable and the objective variable of the plurality of groups. Based on the analysis means for executing the machine learning analysis for each grouping, the values of the explanatory variables of the plurality of groups, and the relationship, the predicted values that are the values of the objective variables of the plurality of groups Prediction means for executing calculation for each grouping, and calculation means for calculating a score relating to the analysis target by calculation based on the predicted value of the group to which the analysis target belongs, calculated for each grouping And comprising.

  The analysis method according to an aspect of the present invention provides a relationship between an explanatory variable and an objective variable of the plurality of groups, for a plurality of groups generated by grouping a plurality of analysis targets in which the explanatory variable and the objective variable are associated. Machine learning analysis is performed for each grouping, and based on the explanatory variable values of the plurality of groups and the relationship, calculation of a predicted value that is the value of the target variable of the plurality of groups, The process is executed for each grouping, and a score related to the analysis target is calculated by calculation based on the predicted value of the group to which the analysis target belongs, calculated for each grouping.

  A program according to an aspect of the present invention provides an explanatory variable and an objective variable of the plurality of groups for a plurality of groups generated by grouping a plurality of analysis targets in which the explanatory variable and the objective variable are associated with each other. Based on the analysis process for performing the machine learning analysis for deriving the relationship for each grouping, the value of the explanatory variable of the plurality of groups, and the relationship, and the prediction that is the value of the objective variable of the plurality of groups The calculation of the value is calculated for each grouping, and the score related to the analysis target is calculated by the calculation based on the predicted value of the group to which the analysis target belongs, calculated for each grouping. And a calculation process.

  According to the present invention, it is possible to perform data analysis in which the influence of the deviation in the number of samples is reduced without reducing or adding sample data.

It is a figure of the data structure showing the example of sample data. It is a block diagram which shows the structure of the analyzer which concerns on the 1st Embodiment of this invention. It is a flowchart which shows the flow of the main operation | movement of the analyzer which concerns on 1st Embodiment. It is a block diagram which shows the structure of the analyzer which concerns on the 2nd Embodiment of this invention. It is a figure which shows the example of the occurrence number of an accident of a certain day in each cell. 6 is a histogram showing the distribution of the number of occurrences of data in the example of FIG. It is a histogram showing distribution of the occurrence number of the data over several hundred days of the occurrence number of accidents in one day in each cell. It is a figure showing the concept which creates a group in the vertical direction. It is a histogram showing distribution of the value of the objective variable of a group. It is a figure showing the concept which creates a group in a horizontal direction. It is a figure explaining the example of the value of each objective variable of the group produced | generated by the grouping of the vertical direction and the grouping of a horizontal direction. It is a figure which shows the example of the predicted value of each group, and the example of the score of each cell calculated from the predicted value. It is a flowchart which shows the flow of operation | movement of the analyzer which concerns on 2nd Embodiment. It is a figure showing the 3rd example of the grouping method. It is a figure showing the 4th example of the grouping method. It is a figure showing the 5th example of the grouping method. It is a figure showing the 6th example of the grouping method. It is a figure which shows the example which derives | leads-out the value of each objective variable of the group produced | generated by the grouping method from the cell data and the grouping method. It is a figure which shows the example which calculates the score of each cell by three grouping methods. It is a figure which shows the example of the predicted value of the objective variable of each group produced | generated by five grouping methods. It is a figure which shows the example of the score of each cell calculated by the modification 3. It is a block diagram which shows the example of the hardware which comprises each part of each embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

<< First Embodiment >>
First, a first embodiment of the present invention will be described.

  The analysis apparatus 11 according to the first embodiment handles sample data, which is a data set of explanatory variables and objective variables accumulated for an analysis target. The analysis target is a target for deriving the value of the objective variable or a guide for the value. The analysis target may be identified by an identification number, or may be identified by a position on the screen displayed by the analysis apparatus 11. Each analysis target is associated with a value of an explanatory variable. The analysis device 11 calculates a score that serves as a guideline for the value of the objective variable to be analyzed based on the value of the explanatory variable by a process described later.

  The objective variable is, for example, a variable selected by the user as a variable whose value is to be predicted. The objective variable may be, for example, the number of accidents per day, the number of incidents per week, or the number of ambulances dispatched per day in a certain area. For example, when the user wants to predict the number of accidents that occur in the next day in a certain area, the objective variable may be set to the number of accidents per day in that area. In this case, the analysis target is the area.

  The explanatory variable is a variable that is considered to be a factor that affects the value of the objective variable. For example, if the objective variable is the number of accidents per day in an area, the variables that can be considered as explanatory variables are, for example, traffic volume, car ownership rate, number of bicycles, number of traffic lights, The number of signs, the number of intersections, the number of past accidents, the weather, the average road width, and whether it is a weekday or a holiday.

  FIG. 1 shows an example of sample data. As shown in FIG. 1, the sample data is a set of explanatory variables and objective variables related to the analysis object for each analysis object (area in this example). In the example shown in FIG. 1, the number of accidents on year Y, month Z, and information such as traffic volume and weather on that day are included in the sample data.

  The analysis device 11 acquires sample data related to the analysis target from, for example, a storage device (not shown). Then, the analysis device 11 calculates, for each of the plurality of analysis targets, a score that serves as a guideline for the value of the objective variable based on the explanatory variable associated with the analysis target. Hereinafter, the control structure of the analyzer 11 will be described.

<Configuration>
FIG. 2 is a block diagram illustrating a configuration of the analyzer 11 according to the first embodiment. The analysis device 11 includes an analysis unit 113, a prediction unit 114, and a calculation unit 115.

  The analysis unit 113 performs machine learning analysis for deriving the relationship between the explanatory variable and the objective variable of each group for a plurality of groups generated by grouping the analysis targets. The grouping means that the analysis target is classified into a plurality of groups.

  The grouping may be performed by, for example, associating a number or the like for identifying a group with each analysis target by a grouping unit (not shown). By grouping, for example, 100 analysis objects are associated with any of 10 groups.

  The analysis unit 113 performs machine learning analysis using a group generated by the grouping as a unit of teacher data.

  Specifically, the analysis unit 113 first creates group data for each group based on sample data associated with the analysis target included in the group. Group data is a combination of explanatory variables and objective variables when the group is regarded as one unit.

  For example, the analysis unit 113 integrates the values of objective variables of sample data associated with analysis targets included in the group. To integrate values is to set a representative value based on each value. That is, the analysis unit 113 sets a representative value based on the value of each objective variable of the sample data associated with the analysis target included in the group, and regards the representative value as the value of the objective variable of the group. Specifically, the values are integrated, for example, by summing those values. Alternatively, integrating values may mean calculating an average of those values.

  Similarly, the analysis unit 113 integrates the values of the respective explanatory variables. When integrating the values of explanatory variables not represented by numerical values, the analysis unit 113 may determine a representative value in the group again based on, for example, how to derive the values of the explanatory variables.

  Note that the analysis unit 113 may integrate objective variables (or explanatory variables) for different days separately. Therefore, for example, when there is sample data over several hundred days, the analysis unit 113 can create [number of classifications to be analyzed × several hundred (pieces)] group data.

  Then, the analysis unit 113 performs machine learning analysis using the created group data as teacher data. The machine learning analysis is an analysis for deriving a relationship between explanatory variables and objective variables based on teacher data, for example. For example, in machine learning analysis, a function representing a relationship between explanatory variables and objective variables is derived. Since this derived function is derived recursively, it is a function that predicts the value of the objective variable from the value of the explanatory variable. Hereinafter, a function derived by machine learning analysis is referred to as a “prediction formula”.

  That is, as an example, the analysis unit 113 derives a prediction formula for predicting the value of the objective variable by analysis using the teacher data.

  The analysis unit 113 derives a prediction formula for each of the two or more groupings. For example, the analysis unit 113 first derives one prediction formula for one grouping by machine learning analysis using group data of all groups generated by one grouping as teacher data. Then, the analysis unit 113 further derives another prediction formula by machine learning analysis using group data of all groups generated by another grouping as teacher data.

  For each grouping, the prediction unit 114 calculates a prediction value that is a value of the objective variable for each group based on the prediction formula derived by the analysis unit 113 and the value of the explanatory variable for each group. For example, the prediction unit 114 calculates the value of the objective variable of the group on the next day based on the prediction formula and the explanatory variable of the group on the next day. The value of the group explanatory variable may be entered by the user. The value of the group explanatory variable may be set based on information included in the database 320.

  The calculation unit 115 calculates a value related to the analysis target by an operation based on the value of the objective variable of the group calculated for each grouping. The value calculated by the calculation unit 115 is referred to as “score”. The score is a measure of the value of the objective variable.

  For example, there are a group objective variable value generated by a certain grouping (first grouping) and a group objective variable value generated by a grouping different from the grouping (second grouping). And The calculation unit 115 multiplies the predicted value of the group to which the analysis target belongs in the first grouping by the predicted value of the group to which the analysis target belongs in the second grouping, as the analysis target score. calculate. The calculation unit 115 may calculate the average of the predicted values of the group to which the analysis target belongs as the analysis target score.

<Main operations>
The main operation flow of the analyzer 11 will be described with reference to the flowchart of FIG.

  In step S21, the analysis unit 113 performs, for each grouping, machine learning analysis for deriving the relationship between the explanatory variables of the plurality of groups and the objective variable for the plurality of groups generated by the grouping for the analysis target.

  In step S <b> 22, the prediction unit 114 calculates, for each grouping, a predicted value that is a value of the target variable of the group based on the value of the explanatory variable of the group and the relationship derived by the analysis unit 113.

  In step S23, the calculation unit 115 calculates a score related to the analysis target by calculation based on the predicted value of the group to which the analysis target belongs, calculated for each grouping.

<Effect>
According to the configuration of the first embodiment, it is possible to perform data analysis in which the influence of the deviation of the number of samples for each value of the objective variable is reduced without reducing or adding sample data. The reason for this is that each sample data is grouped together to reduce the deviation in the number of samples for each value of the objective variable in the teacher data. And in this analysis, the sample data has not been reduced or added.

<< Second Embodiment >>
Next, a second embodiment of the present invention will be described.

<Configuration>
FIG. 4 is a block diagram showing the configuration of the analyzer 12 according to the second embodiment.

  The analysis device 12 is connected to a storage device 32 that stores the database 320 so as to be communicable. The analysis device 12 reads information included in the database 320 from the storage device 32. The information read by the analysis device 12 may be read based on, for example, designation by the user.

  The database 320 of this embodiment includes information related to accidents in a certain area. For example, whether the database 320 is the date and location of the accident in the region, as well as daily traffic, weather, rainfall, number of intersections, number of traffic lights, average road width, and weekdays Data over a predetermined period (for example, the past several hundred days) including information such as discrimination of whether it is a holiday is stored.

  The analysis device 12 includes a division unit 111, a grouping unit 112, an analysis unit 113, a prediction unit 114, a calculation unit 115, and an output unit 116.

=== Division Unit 111 ===
The dividing unit 111 identifies the analysis range (that is, the region range). In specifying the range, the dividing unit 111 acquires information specifying the range of the region from the user, for example. The dividing unit 111 may read information specifying a region range from the database 320. The dividing unit 111 may specify the analysis range based on the information specifying the area range.

  The dividing unit 111 divides the analysis range into a plurality of sections. The size of the mesh when dividing may be appropriately selected according to the purpose. For example, the user may specify the mesh size. In this case, for example, when the user inputs information indicating “1 km square” to the analysis device 12, the dividing unit 111 may divide the area with a mesh size of 1 km square. Alternatively, the size of the mesh may be appropriately set by the analysis device 12 according to the size of the specified region and the number of data.

  Hereinafter, each of the sections generated by the division is referred to as a “cell”. The size of one cell may be several tens of meters square or several kilometers square, for example. The shape of the cell need not be a rectangle. It is not necessary that all cells have the same size.

  The cell generated in this way is a score calculation target by the calculation unit 115 of the analysis apparatus 12 of the present embodiment, that is, an analysis target.

  The dividing unit 111 may specify sample data for each cell. That is, the dividing unit 111 may specify a set of objective variable values and explanatory variable values measured so far for each cell. The objective variable is set by the user, for example. As an example, the objective variable is the number of accidents per day. For example, the dividing unit 111 specifies the values of the objective variable and the explanatory variable that are samples based on the data stored in the database 320. For example, when the analysis device 12 performs analysis using the objective variable as the number of accidents per day, the dividing unit 111 is based on accident data recorded so far, which is stored in the database 320. The number of occurrences per day in each cell and the value of the explanatory variable may be calculated.

  FIG. 5 is a diagram illustrating an example of the number of accidents in each cell on a certain day. A variable that represents the number of occurrences of an event in one cell (for example, 1 km square) in a unit period (for example, 1 day), such as the number of occurrences of an accident, is compared with the case where the value takes a value of 1 or less. In some cases, the value of 2 or more is rare. When such a variable is an objective variable, as shown in FIG. 5, there are many cells whose value of the objective variable is 0 or 1, and there are few cells whose value is 2 or more.

  FIG. 6 is a histogram showing the frequency distribution of the value of the objective variable in each cell in the example shown in FIG. As can be seen from FIG. 6, the number of sample data for each value (number of occurrences) of the objective variable is greatly biased.

  The data handled by the analyzer 12 need not be only data for a specific day. The analysis device 12 may handle data for a plurality of days. FIG. 7 is an example of a histogram showing the frequency distribution of the value of the objective variable of each day's data accumulated in the database 320 over several hundred days. In general, since the number of sample data increases by using data for several hundred days, it is considered that the accuracy of the prediction (precision, that is, the small variation in the prediction result) is improved. However, even if the number of sample data increases, the imbalance in the number of sample data for each value of the objective variable is still not improved, so it cannot be said that accuracy (accuracy, that is, closeness to the true value) is improved.

=== Grouping Unit 112 ===
The grouping unit 112 groups cells in the same column into one group. FIG. 8 is a conceptual diagram showing how the grouping unit 112 groups cells in the same column into one group. In the example illustrated in FIG. 8, the grouping unit 112 groups the cells arranged in the vertical direction into five groups A ₁ , A ₂ , A ₃ , A ₄ , and A ₅ so that the cells are arranged in the same group. That is, the grouping unit 112 classifies the cells into five groups.

  FIG. 9 is an example of a histogram showing the frequency distribution of the value of the objective variable in each group. As shown in FIG. 9, compared to the example shown in FIG. 7, the value of the objective variable is distributed over a wide range, and the variation in the number of sample data for each value of the objective variable is suppressed.

Similarly, the grouping unit 112 groups cells having the same horizontal row into one group. That is, the grouping unit 112 classifies each cell into groups B ₁ , B ₂ , B ₃ , B ₄ , and B _{5 as shown} in FIG.

=== Analysis Unit 113 ===
For each grouping, the analysis unit 113 performs machine learning using the group as a unit of teacher data. Specifically, the analysis unit 113 performs machine learning as follows.

  First, the analysis unit 113 acquires teacher data with a group as one unit. That is, the analysis unit 113 acquires the value of the objective variable and the value of the explanatory variable for each group.

The value of the objective variable of the group is, for example, the sum of the objective variables of the cells included in the group. For example, according to FIG. 8, the value of the objective variable of group A ₁ is 0 + 1 + 1 + 1 + 0 = 3. The value of the objective variable of the group may be an average of the objective variables of the cells included in the group.

  FIG. 11 is a diagram illustrating an example of values of respective objective variables of a group generated by grouping cells in the vertical direction and the horizontal direction in the sample example illustrated in FIG. 5.

  The analysis unit 113 calculates the value of the explanatory variable for each group. The value of the explanatory variable of the group may be, for example, the sum of the explanatory variables of the cells included in the group or an average.

  In this way, the analysis unit 113 obtains teacher data (that is, a set of objective variable values and explanatory variable values) of the group as one unit.

  Then, the analysis unit 113 performs machine learning analysis using the obtained teacher data.

  For example, the analysis unit 113 performs machine learning analysis based on, for example, the past several hundred days of data grouped in the vertical direction, and derives one prediction formula. This prediction formula is a formula for predicting the value of the objective variable of the group from the value of the explanatory variable of the group.

  Similarly, the analysis unit 113 performs machine learning analysis based on the data for the past several hundred days of the group grouped in the horizontal direction, and derives another prediction formula. The past data that is the basis of this machine learning analysis may be data in the same period as the data used in the machine learning analysis for the group grouped in the vertical direction, or may be data in a different period. Good.

  Thereby, the prediction formula based on the teacher data with reduced imbalance is calculated.

=== Prediction unit 114 ===
The prediction unit 114 calculates the predicted value of the objective variable of each group based on the prediction formula derived by the analysis unit 113. Specifically, the prediction unit 114 obtains the predicted value of the objective variable of the group by substituting the value of the explanatory variable of each group into the prediction formula. The value of the explanatory variable of each group used at this time is, for example, the actual value or the predicted value of the explanatory variable on the day on which the predicted value is to be calculated. For example, the same value as the previous day may be set for the number of traffic lights and the number of intersections. The value of the number of bicycles possessed may be set to the same value as the previous day or a value that takes into account the rate of change. A reasonable value may be set for the weather from information such as a weather forecast.

  Hereinafter, the predicted value of the objective variable of the group may be simply referred to as “the predicted value of the group”.

=== Calculating Unit 115 ===
The calculation unit 115 calculates a score serving as a guide for the value of the objective variable of each cell based on the predicted value of the group derived by the prediction unit 114.

  For example, the calculation unit 115 calculates, for each cell, a value obtained by multiplying each predicted value of the group including the cell as a score in the cell.

  FIG. 12 is a diagram illustrating an example of calculating the score of each cell from the predicted value of each group calculated by the prediction unit 114. Note that the Arabic numerals attached to the left of the table and the Roman numerals attached to the top of the table are symbols provided for convenience of explanation, and are symbols for identifying rows or columns in the table, respectively.

As illustrated in FIG. 12, the prediction unit 114 calculates [3, 0, 4, 7, ₁ ] as the prediction values of the groups A _{1 to} A ₅ , respectively, and the prediction values of the groups B _{1 to} B ₅ , respectively. [0, 2, 7, 4, 2] is calculated. In this case, for example, the score of the cell corresponding to the first row and column I is [0 × 3] [0], the score of the cell corresponding to the second row and column I is [2 × 3] [6], The score of the cell corresponding to the third row and the fourth column is [7 × 7], which is [49]. However, “×” is an operator representing multiplication.

The method for calculating the score of each cell in the example of the present embodiment can also be expressed as follows. That is, if the cell column number is represented by i (i = 1, 2, 3, 4, 5) from the left and the row number is represented by j (j = 1, 2, 3, 4, 5) from the top, The score of the cell in the j-th row and the i-th column is B _j × A _i .

  The calculation unit 115 calculates the score of each cell as described above, and associates the calculated value with the cell.

=== Output 116 ===
The output unit 116 outputs information based on the score. For example, the output unit 116 outputs information indicating a cell having a large score value among the cells whose scores are calculated as a place where many accidents are predicted to occur.

  For example, the output unit 116 may extract a predetermined number of cells in order from the cell with the largest score value, and display the extracted cells in a manner different from other cells. For example, the output unit 116 may output an image in which the area corresponding to the cell extracted in the map of the analysis target area is emphasized. The output unit 116 may output data associating cells with score values.

<Operation>
The operation flow of the analyzer 12 according to the second embodiment will be described with reference to FIG.

  First, the dividing unit 111 specifies the analysis range (step S91). Then, the dividing unit 111 divides the analysis range into a plurality of cells (step S92).

  Next, the grouping unit 112 performs grouping by grouping the cells arranged in the vertical direction into the same group (step S93). Then, the analysis unit 113 performs machine learning analysis using the vertical group data as teacher data (step S94). Then, the prediction unit 114 calculates the predicted value of each objective variable of the group based on the prediction formula and the value of each explanatory variable of the vertical group (step S95).

  The grouping unit 112 also performs grouping of cells arranged in the horizontal direction as the same group (step S96). The analysis unit 113 performs machine learning analysis using the horizontal group data as teacher data (step S97). The prediction unit 114 calculates the predicted value of each objective variable of the group based on the prediction formula and the value of each explanatory variable of the horizontal group (step S98).

  The order of processing from step S93 to step S98 is not limited to the above example. For example, the processing from step S93 to step S95 and the processing from step S96 to step S98 may be performed in parallel.

  Then, the calculation unit 115 calculates the score of each cell based on the predicted value of the objective variable of the vertical group and the predicted value of the objective variable of the horizontal group (step S99).

  Finally, the output unit 116 outputs information based on the score (step S100).

<Effect>
According to the analysis device 12 according to the second embodiment, it is possible to calculate a score that is a measure of the value of an objective variable to be predicted in a cell generated by dividing the analysis range into a plurality of ranges.

  For example, according to the specific example described above, the analyzer 12 can calculate a score that is a measure of the number of accidents for each area on the next day. In addition, the analysis device 12 can specify an area that is predicted to have a high number of accidents out of the areas in the analysis range. Further, by outputting the identified result, the user can recognize the area as a place where the risk of accident occurrence is high.

  In this analysis, the analysis device 12 does not need to unnecessarily sort out or discard data included in the database 320 that can be used for machine learning analysis. Further, the analysis device 12 does not need to newly generate or add data that is not included in the database 320.

  That is, the analyzer 12 can perform data analysis without reducing or adding sample data.

  In addition to the above, this analysis reduces the effect of sample number bias for each value of the objective variable. The reason for this is that each sample data is grouped together to reduce the deviation in the number of samples for each value of the objective variable in the teacher data. By reducing the bias, the feature of the sample whose objective variable has a low appearance frequency is less likely to be ignored in machine learning analysis.

<< Modification >>
The explanatory variable and the objective variable handled by the analysis device 12 may be any variables as long as they can be machine learning targets. The objective variable may be the number of specific types of incidents or accidents. In addition, the objective variable may be the number of lightning strikes, the number of reported lost items, the number of dead carcasses found, or the number of damaged public objects.

  The analysis target need not be a region. The analysis target may be an intersection, a police box, or a building. The analysis target may be set according to the objective variable.

  In the following, variations on the grouping method and the score calculation method will be introduced.

(Modification 1)
In the description of the second embodiment, the grouping unit 112 performs grouping by grouping cells arranged in the same column or row into the same group, which is referred to as a grouping method (hereinafter referred to as “grouping method”). ) Is not limited to these.

  14C, FIG. 14D, FIG. 14E, and FIG. 14F are examples of grouping methods other than the grouping method shown above (grouping method C, grouping method D, grouping method E, and group, respectively). It is a figure which shows the division method F). 14C to 14F, Arabic numerals attached to the left side and Roman numerals attached to the upper side of each of the tables shown in FIGS. 14C to 14F are symbols attached for convenience of explanation, and are symbols that respectively identify rows or columns in the tables. It is. The vertical grouping method explained in FIG. 8 is called grouping method A, and the horizontal grouping method explained in FIG. 10 is called grouping method B.

Referring to FIG. 14C, for example, the grouping method C, 1 row I column, second row II column 3 line III column, line 4 column IV, line 5 V sequences in but the same group _{C 1.}

  The grouping unit 112 may employ a grouping method such as the grouping methods C to F. For example, the grouping unit 112 may employ the grouping method A for the first grouping and the grouping method C for the second grouping.

FIG. 15 is a diagram illustrating a calculation example of the value of the integrated objective variable of each group generated by the grouping method C. For example, when the cell shown on the left side of FIG. 15 is grouped by the grouping method shown on the right side of FIG. 15, the value of the explanatory variable of group C ₁ is 0 + 0 + 2 + 2 + 0 = 4.

  The analysis unit 113 performs machine learning analysis based on the first grouping and machine learning analysis based on the second grouping, and derives a prediction formula for each. The prediction unit 114 calculates the predicted value of each group based on each prediction formula, and the calculation unit 115 calculates the multiplication value of the group to which each cell belongs as a score. Also by this method, the score of each cell is calculated. The calculated score is based on a calculation formula specific to the cell. The reason is that every cell has a different group combination including that of other cells.

  The grouping unit 112 may use any two grouping methods as long as the two grouping methods have different combinations of groups including cells for each cell.

  The grouping methods A, B, C, D, E, and F described above are designed so that any two cells belong to different groups in any two or more groups. For the six types of groupings designed in this way, the calculation unit 115 may calculate a score corresponding to each cell using data based on two arbitrary groupings.

(Modification 2)
The grouping unit 112 may execute three or more types of grouping. Then, the analysis unit 113 may derive a prediction formula for each grouping performed by the grouping unit 112. The prediction unit 114 may calculate three or more types of prediction values.

  For example, it is assumed that the grouping unit 112 executes the grouping method A, the grouping method B, and the grouping method C among the grouping methods illustrated in FIGS. 8, 10, and 14 </ b> C to 14 </ b> F. The analysis unit 113 derives a prediction formula based on each grouping. It is assumed that the predicted value of each group is calculated by the prediction unit 114 like the value of the data string to which each group name is assigned in FIG. The calculation unit 115 calculates the score of each cell by multiplying the predicted value of the group to which the cell belongs. Then, the results as shown in the upper left table of FIG. 16 are obtained.

  Thus, the analysis device 12 can also calculate the score of each cell by performing analysis using three or more types of groupings. By analyzing by increasing the types of grouping, the number of predicted values used for calculating the score increases, and the score value of each cell varies more greatly. For example, in the score calculation example of the present modification shown in FIG. 16, the score of the cell in the third row and column I is 42, and the score of the cell in the third row and column III is 112. There is a clear difference. This difference is the difference between the two examples in the score calculation example shown in FIG. 12 (the cell score in the third row and column I is 21, the cell in the third row and column III is the difference). The score is much higher than 28). In this way, by increasing the types of grouping and analyzing, the score value of each cell varies more greatly, thereby making it easier to identify a location with a high occurrence risk. This effect is particularly noticeable when the score is calculated by multiplication.

  Further, in the calculation of the score by multiplication, when the predicted value of a specific group in a certain grouping is 0, the scores of the cells included in the group are all 0, so there is a risk of occurrence of an accident in that cell. It becomes clear that there are few.

  Further, by performing analysis based on three or more types of grouping, the risk that a score of a cell with a low occurrence risk is accidentally calculated large can be reduced.

  Note that the grouping unit 112 may employ a grouping method in which the values of the objective variables of the teacher data are as imbalanced as possible. For example, the grouping unit 112 determines whether the variation in the value of the objective variable of the teacher data of the group generated by the performed grouping deviates from a predetermined standard (for example, whether the variance is lower than the predetermined value). Also good. Then, the grouping unit 112 may perform another grouping again when the variation deviates from a predetermined standard. With this configuration, it is possible to perform machine learning analysis in a group based on a grouping in which variations do not deviate from a predetermined standard.

  The grouping method shown in the first and second modifications can be used in addition to the case where the region is mesh-divided into a plurality of cells. For example, the grouping method described above may be used for a plurality of analysis objects that have already been identified.

  In the above example, the number of analysis objects is 5 × 5, but the number of analysis objects is not limited to this. When the number of analysis objects is not the square of an integer, the number of analysis objects constituting a group may be different for each grouping. For example, when there are 30 analysis targets, the grouping unit 112 divides the first grouping into groups each having five analysis targets and the first grouping into six groups having six analysis targets. You may perform 2 groupings and the 3rd grouping divided | segmented into the group whose analysis object is 5 each. Thus, even if the number of objects to be analyzed is not the square of an integer, a plurality of groups and scores can be calculated.

(Modification 3)
With respect to n × n (n is an integer of 2 or more) analysis objects, as shown in the first modification example, “any two analysis objects seem to belong to different groups in any two or more groupings. When “n + 1” can be created, the calculation unit 115 may calculate the score of the analysis target (cell in the following description) by the method described below. Hereinafter, as Modification 3, a configuration in which the calculation unit 115 calculates the cell score by a method different from the above-described method will be described.

  First, it is assumed that the dividing unit 111 divides an area into n × n cells.

  The grouping unit 112 executes n + 1 types of grouping methods in which any two cells have different group combinations in any two or more groupings for n × n cells. (For example, when n = 5, six of the above-described grouping methods A to F are examples of the above-described n + 1 types of grouping methods.) In other words, the grouping unit 112 includes any two cells, Any one grouping is performed n + 1 times so that it belongs to the same group.

  The analysis unit 113 performs machine learning analysis based on each grouping and derives each prediction formula.

  The prediction unit 114 calculates a predicted value of each group based on each prediction formula.

Then, the calculation unit 115 calculates the cell score as follows. That is,
・ Calculate the sum of all predicted values of the group to which the cell belongs,
Subtract the sum of the predicted values of the group based on an arbitrary grouping (assumed as S) from the calculated sum,
• Divide the subtracted value by n.

  The sum S may be the same in calculating the score of each cell. Further, the value of the sum S may be an average or median of sums of predicted values of groups based on each of a plurality of arbitrary groupings.

  A specific example will be described below using the example shown in FIG. In the case shown in FIG. 5, n = 5.

  The grouping unit 112 performs the grouping methods A to F shown in FIGS. 8, 10 and 14 </ b> C to 14 </ b> F, and calculates explanatory variables and objective variables for each group.

The analysis unit 113 performs machine learning analysis based on each grouping and derives each prediction formula. The prediction unit 114 calculates a predicted value of each group based on each prediction formula. As a result, it is assumed that the predicted value of the group based on each grouping method becomes a value as shown in FIG. That is, the predicted values [a ₁ , a ₂ , a ₃ , a ₄ , a ₅ ] of the groups A _{1 to} A ₅ based on the grouping method A become [ ₃ , 0, ₄ , 7, 1], and the grouping method The predicted values [b ₁ , b ₂ , b ₃ , b ₄ , b ₅ ] of the groups B _{1 to} B ₅ based on B become [ ₀ , ₂ , ₇ , ₄ , ₂ ], and the group C based on the grouping method C The predicted values [c ₁ , c ₂ , c ₃ , c ₄ , c ₅ ] of _{1 to} C ₅ are [ ₄ , ₃ , ₃ , ₂ , ₃ ], and the group D _{1 to} D ₅ based on the grouping method D The predicted values [d ₁ , d ₂ , d ₃ , d ₄ , d ₅ ] become [ ₂ , ₃ , ₂ , ₃ , ₅ ], and the predicted values [e ₁ of the groups E _{1 to} E ₅ based on the grouping method E , E ₂ , e ₃ , e ₄ , e ₅ ] become [ ₂ , ₄ , ₄ , ₂ , ₃ ], and the predicted values [f ₁ , f ₂ , f of the groups F _{1 to} F ₅ based on the grouping method F _3, f , _{F 5]} is to have become a [4,5,2,1,3].

The calculation unit 115 calculates the score of each cell. In calculating the score, the calculation unit 115 calculates a value of the sum S of predicted values of the group based on an arbitrary grouping method. The value of the sum S is, for example, a ₁ + a ₂ + a ₃ + a ₄ + a ₅ .

  The calculation unit 115 calculates a value obtained by dividing, by n, the value obtained by subtracting S from the sum of all predicted values of the group to which the target cell belongs as the score value of the target cell.

For example, since the cell corresponding to 1 row and I column belongs to the groups A ₁ , B ₁ , C ₁ , D ₁ , E ₁ , F ₁ , the score of this cell is
{(A ₁ + b ₁ + c ₁ + d ₁ + e ₁ + f ₁ ) −S} / 5
Is calculated by

Similarly, for example, the score of a cell corresponding to 3 rows and IV columns is
{(A ₄ + b ₃ + c ₂ + d ₅ + e ₃ + f ₁ ) −S} / 5
Is calculated by

  FIG. 18 is a diagram illustrating the score value of each cell calculated by the above method based on the predicted value shown in FIG.

  The score value thus calculated can be regarded as a predicted value of the objective variable of the cell. The reason for this is that the total sum of all predicted values of the group to which the target cell belongs is the sum of the value contributed by all cells one by one and the value contributed by n target cells. Because it can be considered.

  When the objective variable of the group used for the teacher data is calculated by averaging the objective variables of each cell, the step of dividing by n may be omitted in the above-described score calculation step.

  With the method described above, the analysis device 12 can calculate a value with higher accuracy as the predicted value of the objective variable.

(About hardware)
As described above, in each embodiment of the present invention described above, each component of each device represents a functional unit block. A part or all of each component of each device is realized by a possible combination of a computer 1900 and a program as shown in FIG. 19, for example. The computer 1900 includes the following configuration as an example.

CPU (Central Processing Unit) 1901
ROM (Read Only Memory) 1902
RAM (Random Access Memory) 1903
A program 1904A and storage information 1904B loaded into the RAM 1903
A storage device 1905 for storing the program 1904A and storage information 1904B
A drive device 1907 that reads and writes the recording medium 1906
A communication interface 1908 connected to the communication network 1909
Input / output interface 1910 for inputting / outputting data
-Bus 1911 connecting each component
Each component of each device in each embodiment is realized by the CPU 1901 loading the program 1904A for realizing these functions into the RAM 1903 and executing it. A program 1904A for realizing the function of each component of each device is stored in advance in, for example, the storage device 1905 or the ROM 1902, and is read out by the CPU 1901 as necessary. Note that the program 1904A may be supplied to the CPU 1901 via the communication network 1909, or may be stored in advance in the recording medium 1906, and the drive device 1907 may read the program and supply it to the CPU 1901.

  There are various modifications to the method of realizing each device. For example, each device may be realized by a possible combination of a separate computer 1900 and a program for each component. A plurality of components included in each device may be realized by a possible combination of one computer 1900 and a program.

  In addition, some or all of the constituent elements of each device are realized by other general-purpose or dedicated circuits, computers, or combinations thereof. These may be configured by a single chip or may be configured by a plurality of chips connected via a bus.

  When some or all of the constituent elements of each device are realized by a plurality of computers, circuits, etc., the plurality of computers, circuits, etc. may be centrally arranged or distributedly arranged. For example, the computer, the circuit, and the like may be realized as a form in which each is connected via a communication network, such as a client and server system and a cloud computing system.

  The present invention is not limited to the embodiment described above. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

  This application claims the priority on the basis of Japanese application Japanese Patent Application No. 2006-122843 for which it applied on June 21, 2016, and takes in those the indications of all here.

  A part or all of the above embodiment may be described as in the following supplementary notes, but is not limited thereto.

<< Appendix >>
[Appendix 1]
A machine learning analysis for deriving a relationship between an explanatory variable and an objective variable of the plurality of groups is performed on the plurality of groups generated by the grouping that classifies a plurality of analysis targets associated with the explanatory variable and the objective variable. Analysis means to be executed every time,
Prediction means for executing, for each grouping, calculation of a predicted value that is a value of an objective variable of the plurality of groups based on the values of the explanatory variables of the plurality of groups and the relationship;
Calculating means for calculating a score related to the analysis target by calculation based on the predicted value of the group to which the analysis target belongs, calculated for each grouping;
An analyzer comprising:
[Appendix 2]
Any two of the analysis objects belonging to the same group in the grouping belong to different groups in the other grouping,
The analyzer according to appendix 1.
[Appendix 3]
The analysis means performs the machine learning analysis for each of the three or more groupings,
The prediction means executes the calculation of the predicted value for each of the three or more groupings,
The calculation means calculates the score by multiplying each of the predicted values of the group to which the analysis target belongs, calculated for each of the three or more groupings.
The analyzer according to appendix 1 or 2.
[Appendix 4]
For the grouping, it is determined whether the variation in the value of the objective variable of the group generated by the grouping deviates from a predetermined standard, and the grouping is newly executed when the variation deviates from the predetermined standard. Further comprising grouping means,
The analysis means performs the machine learning analysis on a plurality of groups generated by the grouping determined that the variation does not deviate from a predetermined standard.
The analyzer according to any one of appendix 1 to appendix 3.
[Appendix 5]
For n × n (n is an integer of 2 or more) analysis objects, the grouping is performed n + 1 times so that any two analysis objects belong to the same group in any one grouping. Further comprising a grouping means,
The calculation means calculates each score of the analysis target from the sum of the predicted values calculated for all the groups to which the analysis target belongs, for each group calculated based on any of the groupings. Calculate using a value obtained by subtracting the sum of the predicted values.
The analyzer according to appendix 1.
[Appendix 6]
An output means for displaying a predetermined number of the analysis objects from the analysis object having a higher score in a manner different from the other analysis objects;
The analyzer according to any one of appendix 1 to appendix 5.
[Appendix 7]
A machine learning analysis for deriving a relationship between an explanatory variable and an objective variable of the plurality of groups is performed on the plurality of groups generated by the grouping that classifies a plurality of analysis targets associated with the explanatory variable and the objective variable. Run every
Based on the value of the explanatory variable of the plurality of groups and the relationship, calculation of a predicted value that is the value of the target variable of the plurality of groups is performed for each grouping,
A score related to the analysis target is calculated by calculation based on the predicted value of the group to which the analysis target belongs, calculated for each grouping.
Analysis method.
[Appendix 8]
Any two of the analysis objects belonging to the same group in the grouping belong to different groups in the other grouping,
The analysis method according to attachment 7.
[Appendix 9]
Performing the machine learning analysis for each of the three or more groups;
The calculation of the predicted value is performed for each of the three or more groupings,
The score is calculated by multiplying each of the predicted values of the group to which the analysis target belongs, calculated for each of the three or more groupings.
The analysis method according to appendix 7 or 8.
[Appendix 10]
For the grouping, it is determined whether the variation in the value of the objective variable of the group generated by the grouping deviates from a predetermined standard, and the grouping is newly executed when the variation deviates from the predetermined standard. And
Performing the machine learning analysis on a plurality of groups generated by the grouping determined that the variation does not deviate from a predetermined standard;
The analysis method according to any one of appendix 7 to appendix 9.
[Appendix 11]
For n × n (n is an integer of 2 or more) analysis objects, the grouping is performed n + 1 times so that any two analysis objects belong to the same group in any one grouping. ,
The score of the analysis target is subtracted from the sum of the prediction values calculated for all the groups to which the analysis target belongs, and the sum of the prediction values of each group calculated based on any of the groupings. Using the calculated value,
The analysis method according to attachment 7.
[Appendix 12]
Displaying a predetermined number of the analysis objects from the one with the higher score among the analysis objects in a manner different from the other analysis objects;
The analysis method according to any one of appendix 7 to appendix 11.
[Appendix 13]
On the computer,
A machine learning analysis for deriving a relationship between an explanatory variable and an objective variable of the plurality of groups is performed on the plurality of groups generated by the grouping that classifies a plurality of analysis targets associated with the explanatory variable and the objective variable. An analysis process to be executed every time,
Based on the values of the explanatory variables of the plurality of groups and the relationship, a prediction process for calculating a prediction value that is a value of the objective variable of the plurality of groups for each grouping;
A calculation process for calculating a score related to the analysis target by calculation based on the predicted value of the group to which the analysis target belongs, calculated for each grouping;
A program that executes
[Appendix 14]
Any two of the analysis objects belonging to the same group in the grouping belong to different groups in the other grouping,
The program according to attachment 13.
[Appendix 15]
The analysis process performs the machine learning analysis for each of the three or more groupings,
The prediction process executes calculation of the predicted value for each of the three or more groupings,
The calculation process calculates the score by multiplying each of the predicted values of the group to which the analysis target belongs, calculated for each of the three or more groupings.
The program according to appendix 13 or 14.
[Appendix 16]
On the computer,
For the grouping, it is determined whether the variation in the value of the objective variable of the group generated by the grouping deviates from a predetermined standard, and the grouping is newly executed when the variation deviates from the predetermined standard. Execute the grouping process,
The analysis process performs the machine learning analysis on a plurality of groups generated by the grouping determined that the variation does not deviate from a predetermined standard.
The program according to any one of supplementary note 13 to supplementary note 15.
[Appendix 17]
In a computer, for n × n (n is an integer of 2 or more) analysis target, n + 1 times of the sets such that any two of the analysis targets belong to the same group in any one grouping Execute the grouping process to split,
In the calculation process, each score of the analysis target is calculated based on one of the groupings based on the total of the predicted values calculated for all the groups to which the analysis target belongs. Calculate using a value obtained by subtracting the sum of the predicted values.
The program according to attachment 13.
[Appendix 18]
Causing the computer to execute an output process for displaying a predetermined number of the analysis objects in a manner different from the other analysis objects from the analysis object having the higher score.
The program according to any one of appendix 13 to appendix 17.

11, 12 Analysis device 32 Storage device 111 Dividing unit 112 Grouping unit 113 Analyzing unit 114 Prediction unit 115 Calculation unit 116 Output unit 320 Database 1900 Computer 1901 CPU
1902 ROM
1903 RAM
1904A program 1904B storage information 1905 storage device 1906 recording medium 1907 drive device 1908 communication interface 1909 communication network 1910 input / output interface 1911 bus

Claims (18)

A machine learning analysis for deriving a relationship between an explanatory variable and an objective variable of the plurality of groups is performed on the plurality of groups generated by the grouping that classifies a plurality of analysis targets associated with the explanatory variable and the objective variable. Analysis means to be executed every time,
Prediction means for executing, for each grouping, calculation of a predicted value that is a value of an objective variable of the plurality of groups based on the values of the explanatory variables of the plurality of groups and the relationship;
Calculating means for calculating a score related to the analysis target by calculation based on the predicted value of the group to which the analysis target belongs, calculated for each grouping;
An analyzer comprising: Any two of the analysis objects belonging to the same group in the grouping belong to different groups in the other grouping,
The analyzer according to claim 1. The analysis means performs the machine learning analysis for each of the three or more groupings,
The prediction means executes the calculation of the predicted value for each of the three or more groupings,
The calculation means calculates the score by multiplying each of the predicted values of the group to which the analysis target belongs, calculated for each of the three or more groupings.
The analyzer according to claim 1 or 2. For the grouping, it is determined whether the variation in the value of the objective variable of the group generated by the grouping deviates from a predetermined standard, and the grouping is newly executed when the variation deviates from the predetermined standard. Further comprising grouping means,
The analysis means performs the machine learning analysis on a plurality of groups generated by the grouping determined that the variation does not deviate from a predetermined standard.
The analyzer according to any one of claims 1 to 3. For n × n (n is an integer of 2 or more) analysis objects, the grouping is performed n + 1 times so that any two analysis objects belong to the same group in any one grouping. Further comprising a grouping means,
The calculation means calculates each score of the analysis target from the sum of the predicted values calculated for all the groups to which the analysis target belongs, for each group calculated based on any of the groupings. Calculate using a value obtained by subtracting the sum of the predicted values.
The analyzer according to claim 1. An output means for displaying a predetermined number of the analysis objects from the analysis object having a higher score in a manner different from the other analysis objects;
The analyzer according to any one of claims 1 to 5. A machine learning analysis for deriving a relationship between an explanatory variable and an objective variable of the plurality of groups is performed on the plurality of groups generated by the grouping that classifies a plurality of analysis targets associated with the explanatory variable and the objective variable. Run every
Based on the value of the explanatory variable of the plurality of groups and the relationship, calculation of a predicted value that is the value of the target variable of the plurality of groups is performed for each grouping,
A score related to the analysis target is calculated by calculation based on the predicted value of the group to which the analysis target belongs, calculated for each grouping.
Analysis method. Any two of the analysis objects belonging to the same group in the grouping belong to different groups in the other grouping,
The analysis method according to claim 7. Performing the machine learning analysis for each of the three or more groups;
The calculation of the predicted value is performed for each of the three or more groupings,
The score is calculated by multiplying each of the predicted values of the group to which the analysis target belongs, calculated for each of the three or more groupings.
The analysis method according to claim 7 or 8. For the grouping, it is determined whether the variation in the value of the objective variable of the group generated by the grouping deviates from a predetermined standard, and the grouping is newly executed when the variation deviates from the predetermined standard. And
Performing the machine learning analysis on a plurality of groups generated by the grouping determined that the variation does not deviate from a predetermined standard;
The analysis method according to any one of claims 7 to 9. For n × n (n is an integer of 2 or more) analysis objects, the grouping is performed n + 1 times so that any two analysis objects belong to the same group in any one grouping. ,
The score of the analysis target is subtracted from the sum of the prediction values calculated for all the groups to which the analysis target belongs, and the sum of the prediction values of each group calculated based on any of the groupings. Using the calculated value,
The analysis method according to claim 7. Displaying a predetermined number of the analysis objects from the one with the higher score among the analysis objects in a manner different from the other analysis objects;
The analysis method according to any one of claims 7 to 11. On the computer,
A machine learning analysis for deriving a relationship between an explanatory variable and an objective variable of the plurality of groups is performed on the plurality of groups generated by the grouping that classifies a plurality of analysis targets associated with the explanatory variable and the objective variable. An analysis process to be executed every time,
Based on the values of the explanatory variables of the plurality of groups and the relationship, a prediction process for calculating a prediction value that is a value of the objective variable of the plurality of groups for each grouping;
A calculation process for calculating a score related to the analysis target by calculation based on the predicted value of the group to which the analysis target belongs, calculated for each grouping;
A computer-readable non-transitory storage medium storing a program for executing the program. Any two of the analysis objects belonging to the same group in the grouping belong to different groups in the other grouping,
The storage medium according to claim 13. The analysis process performs the machine learning analysis for each of the three or more groupings,
The prediction process executes calculation of the predicted value for each of the three or more groupings,
The calculation process calculates the score by multiplying each of the predicted values of the group to which the analysis target belongs, calculated for each of the three or more groupings.
The storage medium according to claim 13 or 14. The program is stored in the computer.
For the grouping, it is determined whether the variation in the value of the objective variable of the group generated by the grouping deviates from a predetermined standard, and the grouping is newly executed when the variation deviates from the predetermined standard. Execute the grouping process,
The analysis process performs the machine learning analysis on a plurality of groups generated by the grouping determined that the variation does not deviate from a predetermined standard.
The storage medium according to any one of claims 13 to 15. The program causes the computer to make any two of the analysis objects belong to the same group in any one grouping with respect to n × n (n is an integer of 2 or more) of the analysis objects. a grouping process for performing the grouping of n + 1 times is executed,
In the calculation process, each score of the analysis target is calculated based on one of the groupings based on the total of the predicted values calculated for all the groups to which the analysis target belongs. Calculate using a value obtained by subtracting the sum of the predicted values.
The storage medium according to claim 13. The program causes the computer to execute an output process for displaying a predetermined number of the analysis objects in a manner different from the other analysis objects from the analysis object having the higher score.
The storage medium according to any one of claims 13 to 17.

Images