DATA ANALYSIS APPARATUS, DATA ANALYSIS METHOD, AND DATA ANALYSIS PROGRAM

Abstract

To realize highly accurate predictive analysis without depending upon the amount of data.SOLUTION: A data analyzer comprises a processor executing a program, and a storage device storing the program. The data analyzer executes: acquisition processing which acquires a first statistics model based on distribution of measurement results of a group and a second statistics model based on the distribution of first measurement results about a first sample of a sample number smaller than the sample number of the group; calculation processing which calculates correction information indicating difference between the first statistics model and the second statistics model acquired by the acquisition processing; learning processing which generates a first prediction model, by performing machine learning by using the first measurement result, and first feature amount data corresponding to the first measurement result; and correction processing which corrects a first prediction result outputted by inputting second feature amount data of a second sample different from the first sample to the first prediction model generated by the learning processing, by using the correction information calculated by the calculation processing, and outputs a second prediction result.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a data analysis device, a data analysis method, and a data analysis program for analyzing data. [Background technology]

[0002] When performing predictive analysis using machine learning, a minimum of about 50 samples of data are required for learning. Furthermore, when learning is performed using limited data, the predictive model generated by learning tends to fall into the so-called overfitting, where the model is overly adapted to the data. For example, when predicting future disease risk, if learning is only possible using data from a single facility, it is difficult to eliminate the influence of background information related to the facility data, such as patient attributes and facility characteristics. Furthermore, it is often difficult to perform analysis using data from multiple facilities, and it often takes a long time to achieve the desired amount of data.

[0003] For example, Patent Document 1 discloses a data analysis device that derives a regression equation capable of making predictions with an accuracy that meets the requirements of an application in a short period of time even in a situation where sufficient data required for statistical analysis is not collected. This data analysis device extracts data that is a correct value from the target field for a regression equation derived using a simplified pseudo data set having the same attributes as some of the attributes of data collected in a target field to be predicted, which is an explanatory variable useful for predicting a target variable, and an explanatory variable that corrects the deviation in distribution from the collected data, and the target variable, and extracts an explanatory variable that has a significant difference in distribution between the extracted data and a data set collected in a field different from the target field from the correct value data, and creates a pseudo data set that is useful for predicting the target variable in the target field, which has the same feature amount as the collected data, based on large-scale data, for the explanatory variable that has a significant difference. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2017-102710 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the above-mentioned Patent Document 1, it is necessary to collect large-scale data sets such as web data, sensor data, and large-scale questionnaire data. In addition, in Patent Document 1, it is necessary to generate a simplified pseudo data set from the collected large-scale data set, derive a regression equation using the simplified pseudo data set, and add an explanatory variable that corrects the deviation of the distribution from the collected data. In this way, the data analysis device in Patent Document 1 is premised on the collection of a large-scale data set, and therefore cannot perform data analysis if a large-scale data set cannot be collected.

[0006] On the other hand, when providing predictions as a service, being able to make predictions with a certain level of accuracy from the beginning may be a requirement for collecting data over a long period of time or at great cost, or there may be cases where you want to know a certain range of prediction performance in advance.

[0007] The present invention aims to realize highly accurate predictive analysis that is not dependent on the amount of data. [Means for solving the problem]

[0008] A data analysis apparatus according to one aspect of the invention disclosed in the present application includes a processor that executes a program and a storage device that stores the program, and the processor acquires a first statistical model based on a distribution of actual measurement results of a population, and a second statistical model based on a distribution of the first actual measurement results regarding first samples, the number of samples being smaller than the number of samples in the population. No. 1 Calculating correction information indicating an acquisition process and a difference between the first statistical model and the second statistical model acquired by the acquisition process No. 1 A calculation process, the first actual measurement result, and first feature amount data corresponding to the first actual measurement result. A training data set containing Generate a first prediction model by machine learning using No. 1 A learning process; No. 1 A first prediction result output by inputting second feature amount data of a second sample different from the first sample to a first prediction model generated by a learning process is No. 1 Correction is performed using the correction information calculated by the calculation process, First Amendment Output the results No. 1 and performing a correction process. Effect of the Invention

[0009] According to a representative embodiment of the present invention, it is possible to realize highly accurate predictive analysis that is not dependent on the amount of data. Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiments. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is an explanatory diagram showing an example of learning by a data analysis device. [Diagram 2] FIG. 2 is a block diagram illustrating an example of a hardware configuration of the data analysis apparatus. [Diagram 3] FIG. 3 is an explanatory diagram illustrating an example of the correction information. [Figure 4] FIG. 4 is an explanatory diagram illustrating an example of prediction target data and actual measurement results. [Diagram 5] FIG. 5 is an explanatory diagram illustrating an example of the first prediction result. [Figure 6] FIG. 6 is an explanatory diagram illustrating an example of the second prediction result. [Figure 7] FIG. 7 is a graph showing the prediction accuracy of the second prediction result. [Figure 8] FIG. 8 is a flowchart illustrating a first example of a learning process procedure performed by the data analysis device. [Figure 9] FIG. 9 is a flowchart illustrating a second example of a learning process procedure performed by the data analysis device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] <Study example> FIG. 1 is an explanatory diagram showing an example of learning by a data analysis device. In FIG. 1, learning in which the data analysis device 100 predicts the risk of developing a lifestyle-related disease is taken as an example for explanation. Statistical information 101 is, for example, statistical data showing the nationwide age-specific morbidity rate of lifestyle-related diseases obtained based on large-scale data such as the morbidity rate of lifestyle-related diseases using each of a certain group, for example, adults nationwide, as a sample. A nationwide lifestyle-related disease morbidity distribution 102 (hereinafter, simply referred to as the first morbidity distribution 102) is a histogram (not shown) in which the ages in the statistical information 101 are classified by age group IDs according to age groups, and the morbidity rate for each age group ID is plotted. That is, the horizontal axis is the age group ID, and the vertical axis is the morbidity rate of lifestyle-related diseases. A model generated by curve fitting this histogram is the first statistical model 120.

[0012] In this way, the statistical information 101 is (1) statistically analyzed to obtain the first morbidity distribution 102 and the first statistical model 120. Note that the data analysis device 100 does not need the statistical information 101 or the large-scale data underlying the statistical information 101, and it is sufficient if the first statistical model 120 can be obtained.

[0013] The learning data set 103 is, for example, a combination of learning data and supervised data identified by cases in a certain facility A such as a hospital. In other words, the learning data set 103 is a combination of learning data and supervised data related to cases with a smaller number of samples than the huge number of samples like the statistical information 101. Specifically, for example, the learning data set 103 has, as fields, a case ID 130, an age class ID 131, n (n is an integer equal to or greater than 1) features f1 to fn (when the features f1 to fn are not differentiated, they are simply referred to as feature f), and an incidence 132. A combination of values ​​of the fields in the same row defines an entry of one case.

[0014] Case ID 130 is identification information that uniquely identifies a case. Age class ID 131 is identification information that uniquely identifies an age class. Feature value f is quantitative data that indicates the characteristics of a case. Specifically, for example, feature value f includes blood pressure, blood glucose level, and insulin dosage. Feature value f is learning data. Age may also be treated as a feature value, but in this example, age is treated as an age class ID outside of feature value f. Morbidity 132 is the presence or absence of a lifestyle-related disease identified in a case. Morbidity 132 is correct answer data.

[0015] The learning data set 103 is (2) statistically analyzed to obtain an incidence distribution of a certain facility A (hereinafter, simply referred to as a second incidence distribution 104) and a second statistical model 140. The horizontal axis represents the age group ID, and the vertical axis represents the incidence rate of lifestyle-related diseases. The second statistical model 140 is a model generated by curve fitting the points (cases) of the second incidence distribution 104. The data analysis device 100 may generate the second statistical model 140 from the learning data set 103 by a known method, or may obtain the second statistical model 140 from an external source.

[0016] The data analysis apparatus 100 (3) compares and analyzes the first statistical model 120 and the second statistical model 140 to generate correction information 105. The correction information 105 is, for example, information indicating the difference between the first statistical model 120 and the second statistical model 140 for each age group ID 131 on the horizontal axis.

[0017] The data analysis device 100 generates a first prediction model M1 by (4) machine learning the learning dataset 103. The data analysis device 100 inputs prediction target data 106 into the first prediction model M1 and (5) predicts a first prediction result P1. The prediction target data 106 is, for example, data defined by a case ID 130 of facility B, an age group ID 131, and a feature amount f. The first prediction result P1 is, for example, a probability indicating the risk of developing a lifestyle-related disease (morbidity probability).

[0018] The data analysis device 100 (6) corrects the first prediction result P1 with the correction information 105 and outputs a second prediction result. The first prediction model M1 and the (6) correction are the second prediction model M2. The data analysis device 100 generates a loss function 108 using the second prediction result P2 and the actual measurement result 107, and (7) performs additional learning so that the output of the loss function 108 is minimized, thereby adjusting the weight of the first prediction model. This updates the first prediction model M1. The actual measurement result 107 is the morbidity 132 corresponding to the prediction target data 106.

[0019] Thereafter, each time new prediction target data 106 of another facility is obtained, the data analysis device 100 outputs the first prediction result P1 from the latest first prediction model M1, corrects it using the correction information 105 (6) to output the second prediction result P2, and adjusts the weight of the first prediction model by performing additional learning (7) so that the output of the loss function 108 is minimized. In this way, by using the correction information 105, it is possible to realize highly accurate predictive analysis that is not dependent on the amount of data, even if the number of entries in the learning dataset 103 is small. Furthermore, when prediction target data is additionally obtained, the first prediction model is updated, thereby improving the prediction accuracy of the first prediction result P1.

[0020] Furthermore, the data analysis device 100 may (8) add the prediction target data 106 and the actual measurement result 107 to the learning dataset 103. As a result, (2) statistical analysis is performed again on the learning dataset 103 after the addition, and the second incidence distribution 104 and the second statistical model 140 are updated. The data analysis device 100 updates the correction information 105 by (3) comparing and analyzing the first statistical model 120 and the updated second statistical model 140. By applying the updated correction information 105 in (6) correction, the prediction accuracy of the second prediction result P2 is improved.

[0021] In addition, when (8) addition is performed, the data analysis apparatus 100 may perform at least one of (2) updating the second disease distribution 104 and the second statistical model 140 by re-executing the statistical analysis, and updating the correction information 105.

[0022] <Hardware Configuration Example of Data Analysis Apparatus 100> FIG. 2 is a block diagram showing an example of a hardware configuration of the data analysis apparatus 100. The data analysis apparatus 100 includes a processor 201, a storage device 202, an input device 203, an output device 204, and a communication interface (communication IF) 205. The processor 201, the storage device 202, the input device 203, the output device 204, and the communication IF 205 are connected by a bus 206. The processor 201 controls the data analysis apparatus 100. The storage device 202 serves as a working area for the processor 201. The storage device 202 is a non-transient or temporary recording medium that stores various programs and data. Examples of the storage device 202 include a read only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), and a flash memory. The input device 203 inputs data. Examples of the input device 203 include a keyboard, a mouse, a touch panel, a numeric keypad, a scanner, and a microphone. The output device 204 outputs data. The output device 204 may be, for example, a display, a printer, or a speaker. The communication IF 205 connects to a network and transmits and receives data.

[0023] <Examples of various data> 3 is an explanatory diagram showing an example of the correction information 105. The correction information 105 is correction data 300 for each age class ID 131. When the values ​​of the correction data 300 (called correction values) C1, C2, ..., Cr (r is the value of the age class ID 131) are not distinguished from one another, they are simply referred to as correction values ​​C. The correction value Cr is the difference between the value of the first statistical model 120 and the value of the second statistical model 140 for the value r of the same age class ID 131.

[0024] 4 is an explanatory diagram showing an example of prediction target data 106 and actual measurement results 107. The prediction target data 106 is, for example, learning data identified by a case in a facility B different from a facility A. Specifically, for example, the prediction target data 106 has, as fields, a case ID 130, an age class ID 131, and n (n is an integer equal to or greater than 1) feature quantities f1 to fn. A combination of values ​​of the fields in the same row defines an entry of learning data for one case.

[0025] The actual measurement result 107 is, for example, correct answer data identified by a case in facility B. Specifically, for example, the actual measurement result 107 has fields such as a case ID 130, an age group ID 131, and an affected condition 132. A combination of values ​​of the fields in the same row defines an entry of correct answer data for one case. The prediction target data 106 and the actual measurement result 107 can be added to the learning dataset 103 as an additional learning dataset 400.

[0026] 5 is an explanatory diagram showing an example of a first prediction result P1. The first prediction result P1 is prediction data 500 for each case ID 130. When there is no need to distinguish between values ​​of the prediction data 500 (referred to as first predicted values) P1, P2, ..., Pk (k is the value of the case ID 130), they will simply be referred to as first predicted values ​​P.

[0027] 6 is an explanatory diagram showing an example of the second prediction result P2. The second prediction result P2 is prediction data 600 for each case ID 130. When the values ​​of the prediction data 600 (referred to as second predicted values) P1+C3, P2+C7, ..., Pk+Cr are not distinguished from one another, they are simply referred to as the second predicted value P+C. In this embodiment, the second predicted value P+C is expressed as the addition of the first predicted value P and the correction value C, but is not limited to addition, and may be multiplication, or may be any other calculation other than addition or multiplication as long as the correction value C is added to the first predicted value P.

[0028] 7 is a graph showing the prediction accuracy of the second prediction result P2. In the graph 700, the horizontal axis corresponds to the amount of training data, and the vertical axis is the Brier score, with a lower score indicating better prediction accuracy. A characteristic 701 indicates the prediction accuracy of the first prediction result P1 to which the correction information 105 is not applied, and a characteristic 702 indicates the prediction accuracy of the second prediction result P2 when the correction information 105 is applied to the first prediction result P1. The graph 700 shows that the prediction accuracy of the second prediction result P2 is better than the first prediction result P1.

[0029] <Example of learning process procedure> 8 is a flowchart showing a first learning procedure example by the data analysis apparatus 100. The data analysis apparatus 100 acquires a first disease distribution 102 and a first statistical model 120 as a first statistical analysis result (step S801). In addition, the data analysis apparatus 100 acquires a second disease distribution 104 and a second statistical model 140 as a second statistical analysis result (step S802).

[0030] As shown in (3) Comparative Analysis in Fig. 1, the data analysis apparatus 100 compares the first statistical analysis result and the second statistical analysis result to calculate correction information 105 (step S803). As shown in (4) Comparative Analysis in Fig. 1, the data analysis apparatus 100 performs machine learning using the learning dataset 103 to generate a first prediction model M1 (step S804).

[0031] The data analysis device 100 determines whether the prediction target data 106 has been newly input (step S805). If the prediction target data 106 has been newly input (step S805: Yes), the data analysis device 100 inputs the prediction target data 106 into the latest first prediction model M1 to calculate the first prediction result P1 (step S806). The latest first prediction model M1 is the first prediction model M1 generated by the machine learning of (4) if the additional learning of (7) in FIG. 1 has never been performed, and is the first prediction model M1 updated by the most recently performed additional learning if the additional learning of (7) in FIG. 1 has been performed at least once.

[0032] 1 (6) Correction, the data analysis apparatus 100 corrects the first prediction result P1 using the correction information 105 and outputs it as the second prediction result P2 (step S807). Next, the data analysis apparatus 100 calculates a loss function 108 using the actual measurement result 107 corresponding to the prediction target data 106 and the second prediction result P2 (step S808).

[0033] 1 (7) correction, the data analysis apparatus 100 additionally learns the first prediction model M1 so as to minimize the output of the loss function 108 (step S809), and returns to step S805. That is, every time prediction target data 106 of facility C, which is different from facilities A and B, is newly input (step S805: Yes), steps S806 to S809 are executed, so that a highly accurate second prediction result P2 can be output.

[0034] In step S805, when the prediction target data 106 is no longer input (step S805: No), the learning process ends.

[0035] Fig. 9 is a flowchart showing a learning process procedure example 2 by the data analysis apparatus 100. The learning process procedure example 2 is an example of a case where (8) addition of the additional learning data set 400 in Fig. 1 to the learning data set 103 is executed. Note that the same processes as those in Fig. 8 are given the same step numbers, and their explanations will be omitted.

[0036] In the learning process procedure example 2, after step S809, the data analysis device 100 adds the prediction target data 106 and the actual measurement result 107 (additional learning dataset 400) to the learning dataset 103 (step S910), and returns to step S802. In this case, in step S802, the data analysis device 100 acquires the latest second statistical analysis result (second incidence distribution 104 and second statistical model 140) for the latest learning dataset 103 after the addition of the additional learning dataset 400 (step S802).

[0037] After that, the data analysis device 100 compares the first statistical analysis result (the first morbidity distribution 102 and the first statistical model 120) with the latest second statistical analysis result (the second morbidity distribution 104 and the second statistical model 140) to update the correction information 105 (step S803). In addition, the data analysis device 100 re-learns the first prediction model M1 using the latest second statistical analysis result (step S804).

[0038] As a result, every time prediction target data 106 is newly input (step S805: Yes), the data analysis apparatus 100 executes steps S806 to S809 using the latest first prediction result P1 from the re-learned first prediction model M1 and the latest correction information 105. This makes it possible to output a highly accurate second prediction result P2.

[0039] Generally, machine learning focuses on learning from available data, but the data analysis device 100 of this embodiment utilizes correction information 105 derived from large-scale data as information other than the data, and creates a first prediction model M1 and a second prediction model M2 from a learning dataset 103 that has a smaller number of cases than the large-scale data. Therefore, even in a situation where a sufficient learning dataset required for machine learning cannot be collected, predictions can be made with an accuracy that satisfies the application introduction requirements.

[0040] Furthermore, the data analysis apparatus 100 according to the above-described embodiment can also be configured as follows (1) to (5).

[0041] (1) The data analysis device 100 includes a processor 201 that executes a program and a storage device 202 that stores the program. The processor 201 performs an acquisition process (steps S801 and S802) to acquire a first statistical model 120 based on a distribution (first incidence distribution 102) of actual measurement results (statistical information 101) of a population (for example, adults nationwide) and a second statistical model 140 based on a distribution (second incidence distribution 104) of first actual measurement results (incidence 132 of the learning dataset 103 of facility A) for a first sample (case ID 130 of the learning dataset 103 of facility A) whose number of samples is smaller than the number of samples in the population, a calculation process (step S803) to calculate correction information 105 indicating a difference between the first statistical model 120 and the second statistical model 140 acquired by the acquisition process, and a calculation process (step S804) to calculate the correction information 105 indicating a difference between the first actual measurement results (incidence 132 of the learning dataset 103 of facility A) and the first actual measurement results ( A learning process is executed to generate a first prediction model M1 by machine learning using first feature data (feature f of the learning dataset 103 of facility A) corresponding to the first sample (case ID 132 of the learning dataset 103 of facility A) (step S804), and a correction process is executed to correct a first prediction result P1 output by inputting second feature data (feature f of the prediction target data 106 of facility B) of a second sample (case ID 130 of the prediction target data 106 of facility B) different from the first sample (case ID 130 of the learning dataset 103 of facility A) into the first prediction model M1 generated by the learning process, using the correction information 105 calculated by the calculation process, and output a second prediction result P2 (steps S806, S807).

[0042] (2) In the data analysis device 100 of (1) above, the processor 201 executes an additional learning process (step S809) for updating the first prediction model M1 by additional learning using a loss function 108 based on the second prediction result P2 output by the correction process and the second actual measurement result 107 regarding the second sample (case ID 130 of the prediction target data 106 of facility B). In the correction process (steps S806, S807), the processor 201 corrects the third prediction result P1 output by inputting third feature data of a third sample different from the first sample and the second sample (feature f of the prediction target data 106 of facility C) into the first prediction model M1 updated by the additional learning process, using the correction information 105, and outputs a fourth prediction result P2.

[0043] (3) In the data analysis device 100 of (1) above, in the acquisition process, the processor 201 acquires an updated second statistical model 140 based on the distribution of the first actual measurement result (morbidity 132 of the learning dataset 103 of facility A) and the second actual measurement result 107. In the calculation process, the processor 201 calculates updated correction information 105 indicating a difference between the first statistical model 120 and the updated second statistical model 140. In the correction process, the processor 201 corrects a third prediction result (first prediction result P1 for the prediction target data 106 of facility C) output by inputting third feature data (feature f of the prediction target data 106 of facility C) of a third sample different from the first sample and the second sample into the first prediction model M1, using the updated correction information 105, and outputs a fourth prediction result (second prediction result P2 for the prediction target data 106 of facility C).

[0044] (4) In the data analysis device 100 of (1) above, in the acquisition process, the processor 201 acquires an updated second statistical model (second statistical model 140 for facilities A and B) based on the first actual measurement result (morbidity 132 in the learning dataset 103 of facility A) and the distribution of the second actual measurement result 107 (second morbidity distribution 104 for facilities A and B to which the learning dataset 400 of facility B has been added), and in the learning process, the processor 201 updates the first actual measurement result (morbidity 132 in the learning dataset 103 of facility A), the second actual measurement result 107, and the first feature data (feature f) and the second feature data (feature f of the prediction target data 106 of facility B), to generate an updated first prediction model M1, and in the correction process, the processor 201 corrects a third prediction result (first prediction result P1 for the prediction target data 106 of facility C) output by inputting third feature data of a third sample different from the first sample and the second sample (feature f of the prediction target data 106 of facility C) into the updated first prediction model M1 using the correction information 105, and outputs a fourth prediction result (second prediction result P2 for the prediction target data 106 of facility C).

[0045] (5) In the data analysis device 100 of (3) above, in the learning process, the processor 201 performs machine learning again using the first actual measurement result (morbidity 132 of the learning dataset 103 of facility A), the second actual measurement result 107, the first feature data (feature f of the learning dataset 103 of facility A), and the second feature data (feature f of the prediction target data 106 of facility B) to generate an updated first prediction model M1, and in the correction process, the processor 201 corrects a third prediction result (first prediction result P1 for the prediction target data 106 of facility C) output by inputting third feature data of a third sample different from the first sample and the second sample (feature f of the prediction target data 106 of facility C) into the updated first prediction model M1 using the updated correction information 105, and outputs a fourth prediction result (second prediction result P2 for the prediction target data 106 of facility C).

[0046] The present invention is not limited to the above-described embodiments, and includes various modified examples and equivalent configurations within the spirit of the appended claims. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to those having all of the configurations described. Also, a part of the configuration of one embodiment may be replaced with a configuration of another embodiment. Also, a configuration of another embodiment may be added to a configuration of one embodiment. Also, a part of the configuration of each embodiment may be added, deleted, or replaced with another configuration.

[0047] In addition, each of the configurations, functions, processing units, processing means, etc. described above may be realized in hardware, for example by designing some or all of them as an integrated circuit, or may be realized in software by processor 201 interpreting and executing a program that realizes each function.

[0048] Information such as programs, tables, files, etc. that realize each function can be stored in a storage device such as a memory, a hard disk, or an SSD (Solid State Drive), or in a recording medium such as an IC (Integrated Circuit) card, an SD card, or a DVD (Digital Versatile Disc).

[0049] In addition, the control lines and information lines shown are those considered necessary for the explanation, and do not necessarily show all the control lines and information lines necessary for implementation. In reality, it can be considered that almost all components are connected to each other. [Explanation of symbols]

[0050] 100 Data Analysis Equipment 101 Statistics 102 Disease distribution 103 Training Data Set 104 First disease distribution 105 Correction Information 106 Data to be predicted 107 Actual measurement results 108 Loss Function 120 First Statistical Model 130 Case ID 131 Age Class ID 132 Morbidity 140 Second Statistical Model 201 Processor 202 Storage Devices 400 additional training datasets f1~fn feature quantity M1 First prediction model M2 Second forecast model P1 First prediction result P2 2nd prediction result

Claims

1. A data analysis apparatus having a processor that executes a program and a storage device that stores the program, The processor, a first acquisition process for acquiring a first statistical model based on a distribution of actual measurement results of a population and a second statistical model based on a distribution of first actual measurement results regarding first samples, the number of samples being smaller than the number of samples of the population; a first calculation process for calculating correction information indicating a difference between the first statistical model and the second statistical model acquired by the first acquisition process; a first learning process for generating a first prediction model by machine learning using a learning data set including the first actual measurement results and first feature amount data corresponding to the first actual measurement results; a first correction process for correcting a first prediction result output by inputting second feature amount data of a second sample different from the first sample to a first prediction model generated by the first learning process, using the correction information calculated by the first calculation process, and outputting a first corrected result; A data analysis apparatus comprising:

2. The data analysis apparatus according to claim 1 , The processor, an additional learning process for updating the first prediction model by additional learning using a loss function based on a first correction result output by the first correction process and a second actual measurement result regarding the second sample; a second correction process for correcting a second prediction result output by inputting third feature amount data of a third sample different from the first sample and the second sample to a first prediction model updated by the additional learning process, using the correction information, and outputting a second corrected result; A data analysis apparatus comprising:

3. The data analysis apparatus according to claim 1 , The processor, an addition process of adding a second measurement result regarding the second sample and second feature amount data of the second sample to the learning data set; A second acquisition process for acquiring an updated second statistical model updated for the learning data set after the addition by the addition process; a second calculation process for calculating updated correction information indicating a difference between the first statistical model and a second statistical model updated by the second acquisition process; a third correction process for correcting a third prediction result output by inputting third feature amount data of a third sample different from the first sample and the second sample to the first prediction model, using the updated correction information calculated by the second calculation process, and outputting a third corrected result; A data analysis apparatus comprising:

4. The data analysis apparatus according to claim 1 , The processor, an addition process of adding a second measurement result regarding the second sample and second feature amount data of the second sample to the learning data set; A second learning process for updating the first prediction model by performing machine learning using the learning data set after the addition process; a fourth correction process for correcting a fourth prediction result output by inputting third feature amount data of a third sample different from the first sample and the second sample to the first prediction model updated by the second learning process, using the correction information, and outputting a fourth corrected result; A data analysis apparatus comprising:

5. The data analysis apparatus according to claim 3, The processor, A second learning process for updating the first prediction model by performing machine learning using the learning data set after the addition process; In the third correction process, the processor corrects the third prediction result using the updated correction information for the first prediction model updated by the second learning process, and outputs a third correction result. A data analysis apparatus comprising:

6. A data analysis method using a data analysis apparatus having a processor that executes a program and a storage device that stores the program, The processor, a first acquisition process for acquiring a first statistical model based on a distribution of actual measurement results of a population and a second statistical model based on a distribution of first actual measurement results regarding first samples, the number of samples being smaller than the number of samples of the population; a first calculation process for calculating correction information indicating a difference between the first statistical model and the second statistical model acquired by the first acquisition process; a first learning process for generating a first prediction model by machine learning using a learning data set including the first actual measurement results and first feature amount data corresponding to the first actual measurement results; a first correction process for correcting a first prediction result output by inputting second feature amount data of a second sample different from the first sample to a first prediction model generated by the first learning process, using the correction information calculated by the first calculation process, and outputting a first corrected result; A data analysis method comprising:

7. The processor: a first acquisition process for acquiring a first statistical model based on a distribution of actual measurement results of a population and a second statistical model based on a distribution of first actual measurement results regarding first samples, the number of samples being smaller than the number of samples of the population; a first calculation process for calculating correction information indicating a difference between the first statistical model and the second statistical model acquired by the first acquisition process; a first learning process for generating a first prediction model by machine learning using a learning data set including the first actual measurement results and first feature amount data corresponding to the first actual measurement results; a first correction process for correcting a first prediction result output by inputting second feature amount data of a second sample different from the first sample to a first prediction model generated by the first learning process, using the correction information calculated by the first calculation process, and outputting a first corrected result; A data analysis program characterized by executing the above.

Images