JP2001117904A - Predictive diagnostic device for predicting device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent the number of records from being scattered between the threshold values of the result predicted by a predicting device and to directly effectively utilize a predictive diagnostic device for business. SOLUTION: This device is provided with an output part for displaying a predictive diagnostic graph by preparing a certainty table concerning plural predictive values and respective certainties having the intervals of cumulative values for the number of random records generated by the predicting device equipped with a predictive value generating part and a certainty generating part concerning the specified attribute of a new record while using a lot of records containing plural kinds of attribute information, a means for setting the cumulative value for the ideal number of records having fixed intervals concerning the cumulative value for the number of records, and a means for determining the threshold value of the certainty corresponding to the set cumulative value for the ideal number of records.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a prediction diagnosis device for a prediction device.

[0002] A prediction device uses a large amount of past (known) data (records) to perform various predictions on attributes (categories) of unknown data. A device that analyzes what can be understood from the contents using the result predicted by the prediction device is a prediction diagnosis device of the prediction device.

[0003] The prediction device predicts a designated item using data that is close to known data for each record data, and provides a value indicating a certainty factor indicating a degree close to the predicted value and the actual value. Is used to generate a prediction result including the total number of records for each predicted value and to generate a predicted diagnosis graph including the cumulative value of the number of records for each predicted value.

[0004]

2. Description of the Related Art FIG. 7 shows a prediction result table by a conventional prediction diagnosis apparatus. The columns ae in this table are:
Data of records related to a large number of persons, a represents the number of a record corresponding to each different person (in this example, only the numbers 0 to 4 are shown in the row direction), and b represents the age of the person corresponding to each record. , C represent occupations, d represents gender, and e represents actual (known) values for the person of each number. In this example, the actual value e is either one of two values, 1 or 2. For example, an event represented by a binary value such as whether or not a driver's license is obtained or whether or not a smoker is used. It is shown to which of the categories belongs, but there is also an example in which any one of many categories of three or more values is taken.

[0005] f to h in FIG. 7 are data of prediction results.
f is a predicted value based on data of a known record similar to each person, and in this example, is a predicted value related to the same category as that shown in e, and takes either 1 or 2. g is a predicted value 1 representing a probability that the predicted value indicated by f becomes 1
And h is the certainty of the predicted value 2 representing the probability that the predicted value indicated by f becomes 2. For example, in the case of record number 1, the age 30, the housewife, the actual value is 2, the confidence of predicted value 1 is 0.2, and the confidence of predicted value 2 is 0.8,
The predicted value is 2 (the highest confidence value of the predicted value), and the actual value at this time is also 2.

FIG. 8 is a certainty threshold table of the conventional predictive diagnosis apparatus. When a prediction result obtained by the prediction device as shown in FIG. 7 is obtained, the threshold of the certainty factor of the predicted value is determined to a certain size, and the number of records corresponding to the determined threshold is calculated as shown in FIG. And a confidence threshold table representing the accumulation of the number of correct answers and the number of correct answers (the number of actual values that match the predicted value) is created. In this table, when the certainty threshold is 0.9, the accumulation of the number of records having the certainty of 0.9 or more is 10, and the accumulation of the number of correct answers (in this case, the predicted value 1 is the correct answer) is It is one. If the confidence threshold is 0.8
In the above case, the cumulative number of records (including 0.9 or more) is 15, and the cumulative number of correct answers is 5. Hereinafter, similarly, the numerical values of the accumulation of the number of records and the accumulation of the number of correct answers are obtained corresponding to the threshold of each certainty. When the threshold of the certainty is set to 0, the accumulation of the number of records is 100 (the total number of records). , The number of correct answers (predicted value 1) is 40.

FIG. 9 is a conventional predictive diagnosis graph. This predictive diagnosis graph is performed using the certainty threshold table shown in FIG. 8 and is expressed using a two-dimensional graph. The horizontal axis represents the accumulation (total) of the number of records, and the maximum value is the maximum value in FIG. 8 corresponds to the accumulation (= 100), the vertical axis represents the accumulation of the number of correct answers, and the maximum value corresponds to the accumulation (= 40) corresponding to the example of FIG. The dotted line connecting the origin and the diagonal of the graph in FIG. 9 represents the number of correct answers corresponding to the cumulative number of records when randomly obtained without using the predictive diagnosis graph, and the solid line curve represents the curve in FIG. The number of correct answers (predicted value 1) corresponding to each certainty threshold shown in FIG. 7 is plotted on a graph and connected. Also, the number of thresholds of the certainty factor is specified by the user, and the interval is determined by equally dividing [0, 1]. For this reason, the number of records at the threshold value is sparse, and the cumulative value of the number of records in the X-axis direction in the predictive diagnosis graph is biased (the numerical value jumps). For example,
When sending a certain number of direct mails, when predicting so that the degree of certainty of the predicted value that is the correct answer from the target among a large number of target persons reaches a desired number or more, the confidence level of the predicted value reaches a desired value or more. If there is a bias in the accumulated value, an error occurs for a value intermediate between the accumulated values.

[0008]

The conventional predictive diagnosis graph shown in FIG. 9 described above is an analysis of only the information on the number of records and the number of correct answers, and the actual business (cost, income, profit Etc.) cannot be used for the cost evaluation of prediction.

In the determination of the certainty threshold equally, the number of records (the number of samples or the number of data) between the thresholds becomes sparse, and a smooth curve cannot be created in the predictive diagnosis graph. There is a problem that difficult diagnosis becomes difficult. Specifically, for example, assuming an example in which a certain number of direct mails are sent out, it is desired that, from among a large number of target persons, a certainty value of a predicted value that is a correct answer to a purpose be accumulated to a predetermined value or more. When the prediction is performed to reach the number, if the intervals are sparse, the prediction result cannot be accurately diagnosed.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a predictive diagnostic apparatus which can directly use a predictive diagnostic apparatus of a predictive apparatus in business and which prevents the number of records between thresholds of predictive results from being sparse.

[0011]

FIG. 1 shows a first principle configuration of the present invention, and shows a principle configuration for creating a certainty factor corresponding to the ideal number of records by a predictive diagnosis apparatus. Is a configuration diagram. Is an example of a certainty threshold table according to the first principle.

FIG. Where 1 is a prediction device, 1a
Is a predicted value generation unit that generates a predicted value for a newly generated record using known data, 1b is a certainty factor generation unit that generates a certainty factor of the predicted value, 2 is a predictive diagnosis device, 2a
Is a confidence threshold decision unit, 2b is a predictive diagnosis graph output unit, 3
Is a table storage unit, 3a is a certainty threshold table, 3b is a certainty threshold table corresponding to the accumulation of the ideal number of records,
Reference numeral 4 denotes an input unit, and reference numeral 5 denotes an output unit for performing display, printing, and the like.

Conventionally, the threshold of certainty is set at equal intervals, and the accumulation of the number of records and the accumulation of the number of correct answers are obtained for the threshold. However, in the first principle configuration of the present invention, the accumulation of the number of records is performed. Are set equally, and this is set as the number of ideal records. A new threshold of certainty is determined in accordance with each ideal record number, a certainty threshold table is created, and a prediction diagnosis graph is displayed.

FIG. In, a predicted value is generated for each record by the predicted value generation unit 1a of the prediction device 1,
A certainty factor is generated for each predicted value in the certainty factor generating unit 1b, and a predetermined predicted value is used as a correct answer. When a confidence level threshold table (see FIG. 8) similar to that of the related art including the attribute of each record, the predicted value, the predicted value confidence level, and the like is obtained, the prediction diagnostic apparatus 2 compares the record with the confidence level threshold. A confidence level threshold table 3a (see (1) in FIG. 1B) similar to the conventional one including the accumulation of the numbers and the accumulation of the number of correct answers is obtained. When a plurality of values having equal intervals are set as the ideal cumulative number of records (accumulation of the number of records corresponding to the certainty threshold) from the input unit 4 in this table, the certainty threshold determination unit 2a For each cumulative value of the ideal number of records, if there is a cumulative number of records for which the error is equal to or less than a certain value from the certainty threshold table 3a, the corresponding certainty threshold is selected and the error is set. When the value exceeds the value, the new confidence threshold is obtained by interpolation using the confidence threshold corresponding to the nearest cumulative record number, and the cumulative record number error is within the set value by setting the cumulative record number. Then, a confidence threshold table 3b (see B. (2) in FIG. 1) corresponding to the ideal cumulative record number in which a new confidence threshold is set is obtained. The contents of the certainty threshold table 3b are output from the predictive diagnosis graph output unit 2b to the output unit 5. By referring to this display content, the certainty threshold for the ideal number of cumulative records at equal intervals can be immediately known. In the table 3b, the accumulation of the number of correct answers corresponding to each ideal cumulative number of records is not obtained.

FIG. 2 shows a second principle configuration of the present invention, showing a configuration for obtaining income (budget), cost, and profit of a record by the predictive diagnosis apparatus. Is a configuration diagram.
Is an example of a certainty threshold table according to the second principle.

In FIG. 2, 1, 1a, 1b, 2, 2
Reference numerals a, 2b, and 3 to 5 are the same as those in FIG.
Denotes a prediction device, 1a denotes a predicted value generation unit, and 1b denotes a certainty factor generation unit. 2 is a predictive diagnosis device, 2a is a certainty threshold determination unit,
2b is a prediction diagnosis graph output unit, 2c is an income cost setting unit, 2d is a budget desired income setting unit, 2e is a maximum profit detection unit, and 2f is a record number detection unit. Reference numeral 3c in the table storage unit 3 is a certainty threshold table to which new items are added according to the second principle configuration of the present invention.

In the second principle configuration of the present invention, a predictive diagnostic graph showing income, cost and related profit can be represented by using a certainty threshold table.

FIG. In the configuration of FIG. of
For the certainty threshold table for the ideal number of cumulative records shown in (2), the input unit 4 sets the cost c per record (the cost required to process one record, which is assumed to be a percentage of a certain income). I do. Using this cost c, the total cost corresponding to the cumulative value of the number of records for each certainty threshold is obtained, and the cumulative value of the number of correct answers is obtained as income. In addition, the profit for each certainty threshold is calculated as income-cost = profit. This result is shown in FIG. The prediction diagnosis table 3c as shown in FIG.
By driving the predictive diagnosis graph output unit 2b, the output unit 5 has a two-dimensional vertical axis representing income (corresponding to the cumulative number of correct answers), cost, and profit, and a horizontal axis representing a budget (corresponding to the cumulative number of records). Is displayed as a graph. Here, when the maximum profit detecting unit 2e is driven, a certainty threshold having the maximum value is detected from the profits for each of the plurality of certainty thresholds in the prediction diagnosis table 3c. When the budget / desired income is set from the input unit 4 to the budget desired income setting unit 2d, the record number detection unit 2d
f searches the prediction diagnosis table 3c to determine the cumulative value of the number of records corresponding to the set income (the number of records to be processed). As a result, the cost of the record c,
For the setting of the budget and desired income, the optimum number of records to be processed can be easily obtained.

[0019]

FIG. 3 is a processing flow for determining a certainty threshold corresponding to the accumulation of the ideal number of records according to the present invention. Corresponds to the function of the certainty threshold determination unit 2a in the first principle configuration shown in FIG. This processing flow is executed in an information processing device (not shown) including a CPU (central processing unit), a memory, a display device (display), an input device (keyboard, mouse, etc.).

FIG. 4 is a specific example of a certainty threshold table by the predictive diagnosis apparatus of the present invention.

When an item of accumulation of the ideal number of records is added to the conventional certainty threshold table, a new certainty threshold is determined based on the accumulation of the ideal number of records, and a graph creation process is started. An accumulation of the number of records is created (S1 in FIG. 3). When determining a new threshold, the minimum error number (Nir: i = 1, 2,..., N) of the number of records (Nir: i = 1, 2,. p) is set (S2 in FIG. 3). As the value of p, for example, “2” can be set.

In the specific example shown in FIG. 4, the threshold value of each certainty factor (represented by Ri, i =
, N), a threshold value table is created in which the cumulative number of records (Nir), which is a random value, is obtained, and the cumulative number of correct answers corresponding to each is obtained. Cumulative number of ideal records (N
id) are provided, and the contents thereof are equally spaced values (10,
20, 30,..., 100) are set.

Thereafter, a certainty threshold is repeatedly obtained until the cumulative number of records of the newly obtained certainty threshold becomes equal to or less than a set error value. First, i = 1 is set (S in FIG. 3).
3), the selection of the threshold of the i (= 1) th certainty factor in the threshold table is started (S4). That is, it is determined whether all the thresholds have been set (S5 in FIG. 3). If all the thresholds have been set, the process ends.
The accumulation (Njr) of the number of records closest to the accumulation of the ideal number of records (Nid) is searched, and the certainty threshold (Rj
Is obtained (S6 in FIG. 3). In the example of FIG. 4, the threshold value of the certainty factor close to the accumulation (Nid) of the ideal number of records is R
Obtained in order from 1 toward Rn.

It is determined whether the difference between the newly obtained threshold cumulative number of records and the ideal number of cumulative records exceeds the set error (p) (S7 in FIG. 3). Is set as the new threshold (Rj) (S8 in FIG. 3). Thereafter, i is set to +1 and the next threshold value is set (S9 in FIG. 3).
And the same processing is repeated.

On the other hand, if the difference between the cumulative number of records and the cumulative number of ideal records is equal to or larger than the set error in S7, the cumulative number of cumulative records of the closest certainty threshold and the cumulative size of the cumulative number of ideal records are determined. (S1 in FIG. 3)
0), a new certainty factor threshold is set by interpolating with the preceding and following certainty factor thresholds (S11 and S12 in FIG. 3).
That is, when the accumulation of the ideal cumulative record is larger, interpolation is performed using the new threshold (Rj) and the threshold of the certainty factor (Rj + 1) later. The new threshold (Rj) and the previous confidence threshold (Rj-1
) To perform interpolation. Subsequently, a new record number accumulation (Njr) is set (S13 in FIG. 3), and the process returns to S7.
It is determined whether the difference between the set number of records (Njr) and the ideal number of records (Njd) exceeds a set error (p), and the same processing as described above is performed.

In the example of FIG. 4, when i = 1, the accumulation of the ideal number of records = 10 (N1d), which is the same as the accumulation of the number of records = 10 (N1r). .9. When i = 2, the accumulation of the ideal number of records = 20 (N2d), and the accumulation of the corresponding number of records = 19 (N3r) has the smallest error (20-19).
= 1), the accumulated number 19 of records is set, and the threshold value of the certainty factor is 0.7 (R3) corresponding to N3r. Other ideal record numbers (N3d = 30, N4d = 40,...,
Nnd = 100) is not shown. In addition,
The accumulation of the number of correct answers in FIG. 4 is the accumulation of the number of records equal to or greater than the newly obtained certainty threshold.

FIG. 5 shows a certainty threshold table of the predictive diagnosis apparatus to which the items of cost, income and profit according to the present invention are added, and can be created by the second principle configuration of the present invention shown in FIG. .

The certainty threshold determining unit 2a of the predictive diagnosis apparatus 2 shown in FIG. 2 includes the cost, income, and profit items added by the processes of the income cost setting unit 2c and the desired budget income setting unit 2d. A degree threshold table is created. That is, the confidence value threshold value determining unit 2a uses the cumulative value of the number of records (the item represented by Nir, the value of 10, 20, 30,... Of the cumulative value of the ideal number of records shown in FIG. 4 or a value close to the value). Are determined, and the accumulation of the number of correct answers (the item represented by Nd) is set. In this table, three items (cost, income, profit) represented by (B) are added, and the ratio to income is set as the cost (represented by c). In this example, c = 0.3 (30% of income is cost). In this case, the cumulative number of records (Nr) = 10
The cost is 0.3 × 10 = 3, and the income at this time is 1 × 1 = 1 because the cumulative number of correct answers = 1.

The cost is 0.3 × 19 = 5.7 for the next accumulation of the number of records (Nr) = 19, and the income at this time is the accumulation of the number of correct answers = 7. It becomes 7.
Hereinafter, although not shown, costs and incomes are similarly obtained for the accumulation of the number of records, and a certainty threshold table including the added items is created.

The certainty threshold table thus created is processed by the processing of the predictive diagnosis graph output unit so as to be displayed as a graph, and can be displayed on a display device. According to the graph, “Income”, “Cost” and “Profit =
"Income-Cost" is displayed in a graph, and in particular, the cumulative value of the number of records at which the maximum profit can be obtained can be easily found.

FIG. 6 shows the configuration of a predictive diagnosis graph according to the present invention. This graph corresponds to the certainty threshold table shown in FIG. 5. The vertical axis represents the cumulative number of correct answers (Nd), the income (or budget), the cost, and the profit. The horizontal axis represents the cumulative number of records. Represents As for the cost, the cost corresponding to the cumulative value of each record number can be known from a cost line represented by a straight line in proportion to the cumulative number of records.
The income is proportional to the cumulative number of correct answers, and is displayed as an income line having a curve characteristic as shown in the figure.

Further, in FIG. 6, the diameter of the circle is changed according to the certainty threshold corresponding to the accumulated value of the number of records.
9 is represented by a large circle, and the cumulative number of records is 1
9 changes corresponding to the threshold value, such as a circle corresponding to the threshold value 0.7. Instead of displaying the certainty factor according to the size of a circle figure, it may be displayed according to the size of a polygonal figure such as a quadrangle or a triangle.

By displaying such a predictive diagnosis graph, when a horizontal line having a height corresponding to the amount of desired income (budget) is drawn, a point intersecting with the income line is determined, and the horizontal axis of the point is determined. From the coordinates, the cumulative number of records to obtain the income is obtained, and the corresponding cost is found from the vertical axis line of that point, the cost line, and the vertical axis coordinate of the intersection, and the remaining cost after subtracting the cost value is obtained. Profit is determined by the length of the vertical axis.

[0034]

According to the present invention, a means for adding information other than the information generated by the prediction device to the prediction diagnosis device is provided, and the information is used to perform a prediction diagnosis directly related to various businesses. By visually displaying the results graphically, it is possible to easily diagnose the prediction results.

[Brief description of the drawings]

FIG. 1 is a diagram showing a first principle configuration of the present invention.

FIG. 2 is a diagram showing a second principle configuration of the present invention.

FIG. 3 is a diagram showing a processing flow for determining a certainty threshold corresponding to the accumulation of an ideal number of records according to the present invention.

FIG. 4 is a diagram showing a specific example of a certainty threshold table by the predictive diagnosis apparatus of the present invention.

FIG. 5 is a diagram illustrating a certainty threshold table of the predictive diagnosis apparatus to which each item of cost, income, and profit is added according to the present invention.

FIG. 6 is a diagram showing a configuration of a predictive diagnosis graph according to the present invention.

FIG. 7 is a diagram showing a prediction result table by a conventional prediction diagnosis device.

FIG. 8 is a diagram showing a certainty threshold table of the conventional predictive diagnosis apparatus.

FIG. 9 is a diagram showing a conventional predictive diagnosis graph.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Prediction apparatus 1a Predicted value generation part 1b Confidence generation part 2 Prediction diagnostic device 2a Confidence threshold determination part 2b Prediction diagnosis graph output part 3 Table storage part 3a Confidence threshold table 3b Confidence threshold corresponding to accumulation of ideal number of records Table 4 Input unit 5 Output unit

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5B049 AA01 AA06 BB00 CC00 DD01 DD05 EE00 EE01 EE03 EE05 FF03 FF09

Claims (4)

[Claims] 1. A method for accumulating the number of random records generated by a prediction device having a predicted value generation unit and a certainty factor generation unit for a specific attribute of a new record by using a large number of records including a plurality of pieces of attribute information. In a predictive diagnosis apparatus having an output unit for creating a confidence table for a plurality of predicted values having value intervals and respective confidences and displaying a predicted diagnosis graph, the cumulative value of the number of records is determined by a predetermined interval. Means to set the cumulative value of the ideal number of records
And a means for determining a certainty factor threshold value corresponding to the set cumulative value of the ideal number of records. 2. A method according to claim 1, wherein said predictive diagnosis apparatus sets a cost and an income for one record, and calculates the set cost and the income and a difference between them for every fixed accumulation of the number of records. Means for displaying
A prediction diagnosis device for a prediction device, wherein a diagnosis result obtained by the prediction device is diagnosed. 3. An output unit for displaying a predictive diagnosis graph from the certainty factor table according to claim 2, wherein the cost and income values set by the cost and income setting means, and income−cost = profit A predictive diagnostic device for a predictive device, wherein a value is divided and displayed visually. 4. An output unit for displaying the predictive diagnosis graph according to claim 2, wherein each confidence factor is represented by using a figure such as a circle or a rectangle, and the size of the figure is determined according to the magnitude of the confidence degree. A prediction diagnosis device for a prediction device, characterized in that it is determined and displayed.