Information processing device, information processing method, and program

Abstract

To provide an evaluation method capable of accurately evaluating a predicted value of a prediction model and applicable to learning models constructed by various machine learning algorithms, not limited to neural networks.SOLUTION: An information processing device comprises an identification part to identify an important explanatory variable from multiple explanatory variables included in learning data used for the creation of a prediction model, an extraction part to extract similar learning data similar to prediction target data from multiple sets of learning data based on the important explanatory variable value of the learning data and on the important explanatory variable value of the prediction target data. and an evaluation part to evaluate the reliability of the predicted value output from the prediction model using the prediction target data as input, based on the similar learning data extracted by the extraction part.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] Conventionally, there is known a technique for outputting a predicted value for new data by utilizing a learning model constructed by machine learning using training data. Patent Document 1 describes a technique for evaluating a prediction result for a new input example by a trained neural network constructed using training examples. In the technique described in Patent Document 1, similar examples are extracted from the training examples based on the similarity between the training examples and the input examples, and the prediction result by the neural network is evaluated using these similar examples. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2006-236367 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the technology described in Patent Document 1, all explanatory variables of the input examples and training examples are used to evaluate the prediction results, and although the more important the explanatory variables are in the evaluation of the prediction results, if there are many unimportant explanatory variables, they may affect the evaluation. For this reason, the technology described in Patent Document 1 sometimes fails to accurately evaluate the prediction results.

[0005] The present invention has been made in consideration of these circumstances, and one of its exemplary purposes is to provide an information processing device, an information processing method, and a program that enable the predicted values ​​of a prediction model to be evaluated with high accuracy. [Means for solving the problem]

[0006] In order to solve the above problem, an information processing device of one embodiment of the present invention includes an identification unit that identifies important explanatory variables from multiple explanatory variables included in training data used to generate a prediction model; an extraction unit that extracts similar training data that is similar to the data to be predicted from the multiple training data based on values ​​of the important explanatory variables of the training data and values ​​of the important explanatory variables of the data to be predicted; and an evaluation unit that evaluates the reliability of a predicted value output from the prediction model using the data to be predicted as input, based on the similar training data extracted by the extraction unit.

[0007] Yet another aspect of the present invention is an information processing method including: identifying important explanatory variables from a plurality of explanatory variables included in training data used to generate a prediction model; extracting similar training data that is similar to the data to be predicted from the plurality of training data based on values ​​of the important explanatory variables in the training data and values ​​of the important explanatory variables in the data to be predicted; and evaluating, based on the similar training data, the reliability of a predicted value output from the prediction model using the data to be predicted as input.

[0008] Yet another aspect of the present invention is a program for causing a computer to identify important explanatory variables from a plurality of explanatory variables included in training data used to generate a prediction model, extract similar training data that is similar to the data to be predicted from the plurality of training data based on values ​​of the important explanatory variables in the training data and values ​​of the important explanatory variables in the data to be predicted, and evaluate, based on the similar training data, the reliability of a predicted value output from the prediction model using the data to be predicted as input.

[0009] Any combination of the above components, and any transformation of the present invention into a method, device, system, recording medium, computer program, etc., are also valid aspects of the present invention. [Effects of the Invention]

[0010] According to the present invention, it is possible to provide an information processing device, an information processing method, and a program that enable the predicted value of a prediction model to be evaluated with high accuracy. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a diagram illustrating a hardware configuration of an information processing apparatus according to an embodiment of the present invention. [Figure 2] FIG. 2 is a functional block diagram of the information processing device according to the embodiment. [Figure 3] FIG. 2 is a functional block diagram of a processing unit according to the embodiment. [Figure 4] FIG. 10 is a diagram illustrating an example of a method in which an extraction unit calculates a similarity of learning data with respect to prediction target data. [Figure 5] 10 is a flowchart illustrating an example of an operation of an information processing device according to an embodiment of the present invention. [Figure 6] 10 is a flowchart illustrating an example of an operation of an information processing device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0012] [Embodiment] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the description of the drawings, the same elements are given the same reference numerals, and duplicated explanations will be omitted as appropriate. Furthermore, the configurations described below are examples and do not limit the scope of the present invention in any way.

[0013] FIG. 1 is a diagram showing the hardware configuration of an information processing device 1 according to an embodiment of the present invention. The information processing device 1 includes a processor 10, a storage device 12, an input device 14 that accepts input operations, and an output device 16 that outputs information. The processor 10 includes a central processing unit (CPU) and a graphical processing unit (GPU). The storage device 12 includes a memory, a hard disk drive (HDD), and a solid state drive (SSD). The input device 14 includes, for example, a keyboard, a touch panel, a mouse, and a microphone. The output device 16 includes, for example, a display, a touch panel, and a speaker.

[0014] 2 is a functional block diagram of an information processing device 1 according to one embodiment of the present invention. The information processing device 1 according to this embodiment includes an input unit 20, a storage unit 22, a processing unit 24, and an output unit .

[0015] The input unit 20 receives various types of information and transmits the information to the processing unit 24. The input unit 20 is realized by the input device 14 included in the information processing device 1.

[0016] The storage unit 22 stores various types of information. For example, the storage unit 22 may store programs for the processing unit 24 to execute various types of information processing, information about the prediction model, training data used to generate the prediction model, prediction target data, etc. The storage unit 22 is realized by the storage device 12 included in the information processing device 1.

[0017] The prediction model according to this embodiment is generated using training data based on various known machine learning algorithms, and may be configured, for example, by a neural network or a regression tree.

[0018] The learning data according to this embodiment includes a plurality of explanatory variables and a response variable. The explanatory variables according to this embodiment include vehicle-related variables, and may indicate, for example, the vehicle model, type, body color, number of drive wheels (two-wheel drive or four-wheel drive), mileage, and years of use. The response variable according to this embodiment indicates the price of the vehicle. Note that the explanatory variables and response variables of the learning data are not limited to vehicle-related variables.

[0019] A prediction model generated using such training data takes prediction target data as input and outputs the price of a vehicle as a predicted value. The prediction target data according to this embodiment includes multiple explanatory variables similar to those of the training data. Therefore, the prediction model according to this embodiment takes information about a vehicle as input and outputs the price of the vehicle as a predicted value.

[0020] The processing unit 24 executes various information processing operations and transmits the results to the storage unit 22 and the output unit 26. The functions of the processing unit 24 are realized by the processor 10 of the information processing device 1 executing a program stored in the storage device 12.

[0021] The output unit 26 outputs various types of information. For example, the output unit 26 may display the results of information processing by the processing unit 24, and more specifically, may display the evaluation results of the reliability of the predicted values ​​output from the prediction model. The output unit 26 is realized by the output device 16 included in the information processing device 1.

[0022] 3 is a functional block diagram of the processing unit 24 according to this embodiment. The processing unit 24 according to this embodiment includes a generation processing unit 240, an evaluation processing unit 260, and a prediction processing unit 280.

[0023] The generation processing unit 240 generates a prediction model using the training data and determines the importance of the explanatory variables that make up the training data based on the generation results. The functions of the generation processing unit 240 are realized by the cooperation of the model generation unit 242 and the determination unit 244.

[0024] The model generation unit 242 generates a prediction model using training data based on various known machine learning algorithms. Information about the generated prediction model is transmitted to the determination unit 244 and the prediction processing unit 280. The configuration of the prediction model generated by the model generation unit 242 is not particularly limited, but in this embodiment, an example will be described in which the prediction model is configured as a regression tree.

[0025] The determination unit 244 determines the importance of each explanatory variable in the training data. The determination unit 244 may determine the importance of each explanatory variable using feature importance calculated based on various known machine learning algorithms. More specifically, the determination unit 244 may determine the importance of each explanatory variable based on feature importance in a regression tree.

[0026] The determination unit 244 may determine the importance of the explanatory variables based on, for example, feature importance defined in CatBoost (registered trademark). For information on how to calculate feature importance, see the following URL for CatBoost (Feature importance - Model analysis | CatBoost). Note that the feature importance may be calculated by the model generation unit 242. https: / / catboost.ai / en / docs / concepts / fstr#fstr__regular-feature-importance

[0027] When the training data is divided by the explanatory variables in the regression tree of the prediction model, the determining unit 244 may determine the importance of the explanatory variables based on the proximity of the objective variables in the divided set of training data (more specifically, Gini impurity based on the distribution of the objective variables). Specifically, the determining unit 244 may determine the importance of the explanatory variables based on the difference in Gini impurity before and after the division.

[0028] For example, suppose that the Gini impurity before and after splitting the training data by a certain explanatory variable A is difference A, and the Gini impurity before and after splitting the training data by a certain explanatory variable B is difference B. The determination unit 244 can determine the importance of explanatory variables A and B by comparing difference A and difference B. Specifically, the determination unit 244 may determine that the greater the decrease in Gini impurity before and after splitting the training data, the higher the importance of the corresponding explanatory variable.

[0029] There may be cases where the training data is divided multiple times for a certain explanatory variable, such as dividing the training data under a first condition for explanatory variable A, then dividing the training data under explanatory variable B, and then dividing the training data under a second condition for explanatory variable A. In this case, the determination unit 244 takes into account the difference in Gini impurity when dividing the training data under the first and second conditions for at least explanatory variable A.

[0030] If the prediction model is configured as a neural network, the determining unit 244 may determine the importance of each explanatory variable based on the connection state of the neural network.

[0031] The evaluation processing unit 260 evaluates the reliability of the predicted value output from the prediction model based on the prediction model generated by the generation processing unit 240 and the determined importance of the explanatory variables. The function of the evaluation processing unit 260 is realized by the identification unit 262, extraction unit 264, counting unit 266, calculation unit 268, and evaluation unit 270 working together.

[0032] The identification unit 262 identifies important explanatory variables from multiple explanatory variables included in the learning data used to generate the prediction model. Specifically, the identification unit 262 may identify important explanatory variables based on the importance of the explanatory variables determined by the determination unit 244. For example, the identification unit 262 may identify explanatory variables having importance equal to or greater than a predetermined threshold as important explanatory variables. The identification unit 262 transmits information about the important explanatory variables to the extraction unit 264.

[0033] The extraction unit 264 extracts similar training data that is similar to the prediction target data from the multiple training data based on the values ​​of the important explanatory variables in the training data and the values ​​of the important explanatory variables in the prediction target data. Specifically, the extraction unit 264 may calculate the similarity of the training data to the prediction target data based on the values ​​of the important explanatory variables in the training data and the values ​​of the important explanatory variables in the prediction target data, and extract similar training data using the similarity. For example, if the similarity of the training data exceeds a predetermined threshold, the extraction unit 264 may extract the training data as similar training data. The extraction unit 264 transmits information about the extracted similar training data to the counting unit 266, the calculation unit 268, and the prediction processing unit 280.

[0034] An example of a method by which the extraction unit 264 calculates the similarity of training data to the data to be predicted will be described with reference to Fig. 4. Fig. 4 shows the values ​​of three important explanatory variables (X1, X2, and X3) for the data to be predicted and one training data. Here, the importance of X1, X2, and X3 is 0.7, 0.2, and 0.1.

[0035] As shown in Figure 4, the values ​​of the important explanatory variables X1, X2, and X3 of the data to be predicted are 0.7, 0.5, and 1, respectively. The values ​​of the important explanatory variables X1, X2, and X3 of the training data are 0.2, 0.1, and 0, respectively. In this example, the similarity is calculated based on the values ​​of these important explanatory variables by performing the following processes (1) to (3).

[0036] (1) Calculate the absolute value of the difference between the values ​​of the important explanatory variables First, the absolute values ​​of the differences between the values ​​of the important explanatory variables in the training data and the values ​​of the important explanatory variables in the data to be predicted are calculated. As shown in FIG. 4, the calculation results are 0.5, 0.4, and 1 for the important explanatory variables X1, X2, and X3, respectively. Note that in this example, an example is described in which similar training data is extracted using the absolute values ​​of the differences between the values ​​of the important explanatory variables in the training data and the values ​​of the important explanatory variables in the data to be predicted. However, for example, the product (also called "cosine similarity") of the values ​​of the important explanatory variables in the training data and the values ​​of the important explanatory variables in the data to be predicted may be calculated, and similar training data may be extracted based on this product.

[0037] Furthermore, when the important explanatory variables are quantitative variables, min-max normalization may be performed before calculating the absolute value of the difference between the values ​​of the important explanatory variables so that the minimum value of the important explanatory variables in all the training data becomes 0 and the maximum value of the important explanatory variables in all the training data becomes 1.

[0038] Furthermore, if the important explanatory variable is a qualitative variable, the value of the important explanatory variable may be set to 1 when a predetermined condition is satisfied, and may be set to 0 when the predetermined condition is not satisfied. For example, if the predetermined condition is that the vehicle color is black, the value of the important explanatory variable may be set to 1 when the vehicle color indicated by the value of the important explanatory variable is black, and may be set to 0 when the vehicle color indicated by the value of the important explanatory variable is a color other than black.

[0039] (2) Multiplying the absolute value and the importance The absolute values ​​of the differences between the values ​​of the important explanatory variables calculated in (1) above and the importance of the important explanatory variables are multiplied. As shown in Figure 4, the multiplication results are 0.35, 0.08, and 0.1 for the important explanatory variables X1, X2, and X3, respectively.

[0040] (3) Calculation of similarity The similarity of the training data is calculated using the sum of the values ​​obtained by the integration in (2) above. Specifically, 0.47, which is the value obtained by subtracting the sum of the integrated values ​​(0.35 + 0.08 + 0.1 = 0.53) from 1, is calculated as the similarity of the training data. In the same manner, the extraction unit 264 can calculate the similarity for all training data, and similar training data is extracted from all training data based on the similarity calculated in this way.

[0041] The method for calculating the similarity of the training data is not limited to the example described here. When the important explanatory variables are quantitative variables, the extraction unit 264 may calculate the similarity based on the distance between the training data and the data to be predicted. Specifically, the extraction unit 264 may calculate the similarity based on the square root of the sum of squares using the differences in the values ​​of the important explanatory variables. When the important explanatory variables are qualitative variables, the extraction unit 264 may calculate the similarity based on the number of important explanatory variables that match between the training data and the data to be predicted. For example, the extraction unit 264 may calculate the similarity so that the larger the number of matching important explanatory variables is, or the larger the ratio of the number of matching important explanatory variables to the number of all important explanatory variables is, the larger the similarity is.

[0042] 3 , the function of the evaluation processing unit 260 will be described. The counting unit 266 of the evaluation processing unit 260 counts the number of similar learning data extracted by the extraction unit 264 and transmits the counting result to the evaluation unit 270. In addition, the calculation unit 268 calculates the standard deviation of the objective variable of the multiple similar learning data extracted by the extraction unit 264 and transmits the calculated standard deviation to the evaluation unit 270.

[0043] The evaluation unit 270 evaluates the reliability of the predicted value output from the prediction model when the prediction target data is input, based on the similar learning data extracted by the extraction unit 264. The evaluation unit 270 may transmit the evaluation result to the output unit 26. As a result, the output unit 26 displays the evaluation result.

[0044] In this embodiment, the evaluation unit 270 evaluates the reliability of the predicted value based on the number of similar training data counted by the counting unit 266. Specifically, the evaluation unit 270 may evaluate the reliability of the predicted value higher as the number of similar training data increases. Alternatively, the evaluation unit 270 may evaluate the reliability of the predicted value higher as the ratio of the number of similar training data to the total number of similar training data increases. It is considered that the more similar training data is used to generate the prediction model, the higher the reliability of the predicted value output from the prediction model. Therefore, by using the number of similar training data in the evaluation, the reliability of the predicted value can be evaluated more accurately.

[0045] Furthermore, the evaluation unit 270 evaluates the reliability of the predicted value based on the standard deviation of the objective variable of the similar learning data calculated by the calculation unit 268. Specifically, the smaller the standard deviation of the objective variable of the similar learning data, the higher the evaluation unit 270 may evaluate the reliability of the predicted value. On the other hand, the larger the standard deviation of the objective variable of the similar learning data, the lower the evaluation unit 270 may evaluate the reliability of the predicted value. The evaluation unit 270 may calculate the standard deviation or a value based on the standard deviation as the reliability.

[0046] The smaller the standard deviation of the objective variable, the smaller the variance in the objective variable values. A predictive model generated using similar training data with objective variable values ​​that have small variances can be expected to output highly accurate predicted values. On the other hand, a predictive model generated using similar training data with objective variable values ​​that have large variances is likely to output less accurate predicted values. Therefore, by using the standard deviation of the objective variable of the similar training data in the evaluation, the reliability of the predicted values ​​can be evaluated more accurately.

[0047] The prediction processing unit 280 generates a predicted value using the prediction model generated by the model generation unit 242, and corrects the predicted value. The function of the prediction processing unit 280 is realized by the prediction generation unit 282 and the correction unit 284 working together.

[0048] The prediction generation unit 282 generates a predicted value for the data to be predicted using the prediction model, and transmits the predicted value to the correction unit 284. Specifically, the prediction generation unit 282 inputs the explanatory variables of the data to be predicted into the prediction model, and obtains the predicted value.

[0049] The correction unit 284 corrects the predicted value generated by the prediction generation unit 282 and transmits the corrected result to the output unit 26. As a result, the predicted value is displayed on the output unit 26. Specifically, the correction unit 284 may correct the predicted value based on a plurality of similar learning data extracted by the extraction unit 264. By correcting the predicted value using similar learning data, it is possible to correct the predicted value more accurately.

[0050] Even for vehicles with the same vehicle make, model, mileage, and other conditions, the value of the vehicle often changes from year to year. Therefore, when there are multiple similar learning data, the values ​​of these objective variables (vehicle prices) may change significantly depending on the year corresponding to each similar learning data. In this embodiment, the correction unit 284 corrects the predicted value taking into account such changes in the objective variable values ​​over time.

[0051] In this embodiment, each of the multiple learning data is associated with time information. The time information may be, for example, information indicating the year and date when the vehicle price corresponding to the value of the objective variable was determined by appraisal and when the vehicle was sold for the price corresponding to the value of the objective variable. The correction unit 284 can correct the predicted value based on the relationship between the value of the objective variable and the time information.

[0052] For example, suppose the value of the objective variable in the training data tends to increase over time. In this case, the price of a vehicle that includes conditions similar to the values ​​of the explanatory variables in the training data based on past information is estimated to be higher than the vehicle price predicted based on the training data. However, because the prediction model is generated using the training data, the time dependency of the objective variable value is not reflected in the prediction model. Therefore, when the data to be predicted is input into the prediction model, there is a possibility that the output price will be lower than the actual vehicle price.

[0053] Therefore, the correction unit 284 may correct the predicted value output from the prediction model based on the time dependency of the value of the objective variable. Specifically, if the value of the objective variable is increasing over time, a price corresponding to the rate at which the value of the objective variable is increasing may be added to the predicted value. Also, if the value of the objective variable is decreasing over time, a price corresponding to the rate at which the value of the objective variable is decreasing may be subtracted from the predicted value. By correcting the predicted value using the time dependency of the value of the objective variable in this way, it becomes possible to correct the predicted value more accurately when the objective variable is something whose price fluctuates from year to year, such as the price of a vehicle.

[0054] 5 is a flowchart showing an example of the operation of the information processing device 1 according to one embodiment of the present invention. The flow of the operation according to this embodiment will be described below with reference to the flowchart shown in FIG.

[0055] First, the model generation unit 242 generates a prediction model using multiple pieces of training data (S101). At this time, feature importance may be calculated along with the generation of the prediction model. Next, the determination unit 244 determines the importance of explanatory variables in the training data (S103). Here, the determination unit 244 determines the importance of multiple explanatory variables included in the training data used to generate the prediction model in S101. Next, the identification unit 262 identifies important explanatory variables based on the importance of the explanatory variables determined in S103 (S105).

[0056] Next, the extraction unit 264 extracts similar training data from the multiple training data used to generate the prediction model in S101 based on the values ​​of the important explanatory variables of the training data and the values ​​of the important explanatory variables of the data to be predicted (S107). The important explanatory variables used here are the important explanatory variables identified in S105. Next, the counting unit 266 counts the number of similar training data extracted in S107 (S109). Next, the calculation unit 268 outputs the standard deviation of the objective variable of the similar training data extracted in S107 (S111).

[0057] Next, the evaluation unit 270 evaluates the predicted value output from the prediction model when the prediction data is input, based on the similar training data (S113). Here, the evaluation unit 270 evaluates the reliability of the predicted value based on the number of similar training data counted in S109 and the standard deviation of the dependent variable calculated in S111. When the prediction unit 252 evaluates the reliability of the predicted value, the evaluation process shown in FIG. 5 ends.

[0058] 6 is a flowchart showing an example of the operation of the information processing device 1 according to one embodiment of the present invention. The flow of the evaluation process according to this embodiment will be described below with reference to the flowchart shown in FIG.

[0059] First, the processes of S201 to S207 are performed. These processes are substantially the same as the processes of S101 to S107 described with reference to FIG. 5, and therefore will not be described here. When similar learning data is extracted in S207, the prediction generation unit 282 uses a prediction model to generate a predicted value using the prediction target data as input (S209). Next, the correction unit 284 corrects the predicted value generated in S209 based on the relationship between the value of the objective variable of the similar learning data extracted in S207 and the time information (S211). When the correction unit 284 corrects the predicted value, the correction process shown in FIG. 6 ends.

[0060] The configuration and operation of an information processing device 1 according to one embodiment of the present invention have been described above. The information processing device 1 according to this embodiment extracts similar learning data that is similar to data to be predicted based on important explanatory variables identified from multiple explanatory variables, and can use the similar learning data to evaluate the reliability of predicted values ​​output by a prediction model. Therefore, even if the learning data includes unimportant explanatory variables, such explanatory variables are not used to evaluate the reliability of the predicted values, allowing for accurate evaluation of predicted values ​​from a prediction model.

[0061] Furthermore, the information processing device 1 according to this embodiment can evaluate the reliability of predicted values ​​for prediction models constructed using not only neural networks but also various other models such as regression trees. Therefore, the information processing device 1 according to this embodiment can provide a highly versatile method for evaluating the reliability of predicted values.

[0062] Furthermore, according to the information processing device 1 of this embodiment, the evaluation unit 270 evaluates the reliability of a predicted value based on the number of similar learning data. Therefore, the evaluation unit 270 can evaluate that the more similar learning data there are, the higher the reliability, and can evaluate the reliability with higher accuracy.

[0063] Furthermore, according to the information processing device 1 of this embodiment, the evaluation unit 270 evaluates the reliability of the predicted value based on the standard deviation of the objective variable of the similar learning data. Therefore, the evaluation unit 270 can evaluate that the reliability of the predicted value is higher when the variance of the objective variable of the similar learning data is smaller, and it is possible to evaluate the reliability more accurately.

[0064] The information processing device 1 according to this embodiment can also be used to evaluate the reliability of the predicted value of a prediction model for predicting the price at which a used vehicle will be sold at auction. The selling price of a used vehicle is generally assessed by an appraiser based on their own intuition, experience, and price experience. According to this embodiment, by using a prediction model generated by the information processing device 1, it is possible to predict the price of a vehicle without relying on the appraiser's intuition, and to evaluate the reliability of the predicted value.

[0065] [supplement] The present invention has been described above based on the embodiments. These embodiments are merely examples, and it will be understood by those skilled in the art that various modifications are possible in the combination of the components and treatment processes, and that such modifications are also within the scope of the present invention.

[0066] The processing of each step by the information processing device 1 described with reference to the above flowchart does not necessarily have to be performed in the order shown. The processing of the steps may be appropriately interchanged within a logically consistent range, or multiple steps may be performed in parallel. For example, the processing of S109 (counting the number of similar training data) and S111 (calculating the standard deviation of the objective variable of similar training data) may be performed in parallel. [Explanation of symbols]

[0067] 1 Information processing device, 240 generation processing unit, 242 model generation unit, 244 judgment unit, 252 prediction unit, 260 evaluation processing unit, 262 identification unit, 264 extraction unit, 266 counting unit, 268 calculation unit, 270 evaluation unit, 282 prediction generation unit, 284 correction unit, 30 regression tree.

Claims

1. an identification unit that identifies important explanatory variables from a plurality of explanatory variables included in the learning data used to generate the prediction model; an extraction unit that extracts similar learning data that is similar to the prediction target data from a plurality of the learning data based on values ​​of important explanatory variables of the learning data and values ​​of important explanatory variables of the prediction target data; an evaluation unit that evaluates the reliability of a predicted value output from the prediction model using the prediction target data as input, based on the similar learning data extracted by the extraction unit, Information processing device.

2. the evaluation unit evaluates the reliability of the predicted value based on the number of similar training data extracted by the extraction unit. The information processing device according to claim 1 .

3. the extraction unit extracts a plurality of the similar training data, each of which is similar to the prediction target data, from the plurality of training data; the evaluation unit evaluates the reliability of the predicted value based on a standard deviation of a dependent variable of the plurality of similar learning data. The information processing device according to claim 1 .

4. the explanatory variables include vehicle-related variables, The objective variable of the learning data indicates the price of the vehicle. The information processing device according to claim 1 .

5. the predictive model includes a regression tree; the identification unit identifies the important explanatory variables based on feature importance in the regression tree. The information processing device according to claim 4 .

6. a prediction generating unit that generates the prediction value by using the prediction model and the prediction target data as input; a correction unit that corrects the predicted value generated by the prediction generation unit based on the similar learning data, the extraction unit extracts a plurality of similar training data items each similar to the prediction target data item from the plurality of training data items; each of the plurality of similar training data is associated with time information; the correction unit corrects the predicted value based on a relationship between the value of the dependent variable and the time information.

6. The information processing device according to claim 4.

7. An information processing method by an information processing device having a processor, comprising: the processor: Identifying important explanatory variables from a plurality of explanatory variables included in the training data used to generate the predictive model; extracting similar learning data that is similar to the prediction target data from the plurality of learning data based on values ​​of important explanatory variables of the learning data and values ​​of important explanatory variables of the prediction target data; and evaluating the reliability of a predicted value output from the prediction model using the prediction target data as input based on the similar learning data. Information processing methods.

8. On the computer, Identifying important explanatory variables from a plurality of explanatory variables included in the training data used to generate the predictive model; extracting similar learning data that is similar to the prediction target data from the plurality of learning data based on values ​​of important explanatory variables of the learning data and values ​​of important explanatory variables of the prediction target data; Evaluating the reliability of a predicted value output from the prediction model using the prediction target data as input based on the similar learning data; A program to execute.

Images