Prediction model creation device, prediction model creation method, and program

Abstract

To solve the problem of a decrease in security when creating a prediction model by machine learning.SOLUTION: A prediction model creation device according to the present disclosure includes an acquisition unit that acquires training data including averaged samples obtained by averaging a plurality of samples each including a set of an explanatory variable and an objective variable, an estimation unit that estimates a pre-averaging distribution which is a distribution of the explanatory variable before averaging corresponding to the explanatory variable constituting the averaged samples which are the training data, and a learning unit that performs machine learning of a prediction model for predicting the objective variable from the explanatory variable based on the training data and the pre-averaging distribution.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present disclosure relates to a predictive model creation device, a predictive model creation method, and a program. [Background technology]

[0002] Predictions based on input data are being made using machine learning models in a variety of fields. For example, Patent Document 1 describes the use of machine learning models that have been trained on past case data to predict loan screening decisions and patient illnesses. As a specific example, Patent Document 1 describes the creation of a machine learning model by machine learning case data such as the patient's gender, age, and medical procedures. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-76345 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the technology described in Patent Document 1, personal information such as the patient's age and the medical procedure performed is used as training data for machine learning, which may result in the leakage of such personal information. For example, there is a risk that personal information may be leaked from the training data itself or from the machine learning model. As a result, when creating a predictive model using case data, there is a risk that raw data containing personal information may be leaked, which may result in a decrease in security.

[0005] Therefore, one of the objectives of the present disclosure is to solve the above-mentioned problem that a decrease in security may occur when creating a predictive model using machine learning. [Means for solving the problem]

[0006] A predictive model creation device according to one embodiment of the present disclosure includes: an acquisition unit that acquires training data consisting of averaged samples obtained by averaging a plurality of samples each consisting of a pair of an explanatory variable and a target variable; an estimation unit that estimates a pre-averaging distribution, which is a distribution of explanatory variables before averaging corresponding to explanatory variables that constitute the averaged samples that are the training data; a learning unit that performs machine learning to generate a prediction model that predicts a dependent variable from an explanatory variable based on the training data and the pre-averaging distribution; Equipped with The structure is as follows. Furthermore, a prediction model creation method according to an embodiment of the present disclosure includes: Obtain training data consisting of averaged samples obtained by averaging a plurality of samples each consisting of a pair of explanatory variables and a target variable; Estimating a pre-averaging distribution, which is a distribution of explanatory variables before averaging corresponding to explanatory variables constituting the averaging sample, which is the training data; a prediction model that predicts a dependent variable from an explanatory variable based on the training data and the averaging prior distribution; The structure is as follows. Furthermore, a program according to an embodiment of the present disclosure includes: Obtain training data consisting of averaged samples obtained by averaging a plurality of samples each consisting of a pair of explanatory variables and a target variable; Estimating a pre-averaging distribution, which is a distribution of explanatory variables before averaging corresponding to explanatory variables constituting the averaging sample, which is the training data; a prediction model that predicts a dependent variable from an explanatory variable based on the training data and the averaging prior distribution; Have the computer perform the process, The structure is as follows. [Effects of the Invention]

[0007] By being configured as described above, the present disclosure can improve security when creating a predictive model using machine learning. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram illustrating an example of a configuration of a predictive model creation device according to the present disclosure. [Figure 2] 10 is a flowchart illustrating an example of a processing operation of a predictive model creation device according to the present disclosure. [Figure 3] 1 is a block diagram illustrating an example of a configuration and processing state of an information processing system according to the present disclosure. [Figure 4] FIG. 1 is a block diagram illustrating an example of a hardware configuration of a predictive model creation device according to the present disclosure. [Figure 5] FIG. 1 is a block diagram illustrating an example of a configuration of a predictive model creation device according to the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0009] First Embodiment A first embodiment of the present disclosure will be described with reference to the drawings, which may be relevant to any embodiment.

[0010] The predictive model creation device 10 in this embodiment performs machine learning on training data to create a predictive model that predicts a dependent variable from an explanatory variable. In particular, in this embodiment, raw data of cases is not used as training data as is, but rather an averaged sample obtained by averaging multiple raw data samples is used as training data to create a predictive model. This makes it possible to prevent the leakage of raw data such as personal information, which is training data, and improves security.

[0011] In this embodiment, as an example of prediction using a prediction model, a case where an individual's loan screening decision prediction is performed will be described. In this case, the explanatory variables input to the prediction model include the individual's age, annual income, savings, place of employment, years of service, etc., and the objective variable output from the prediction model is creditworthiness, such as whether or not the loan will be approved. However, the prediction using the prediction model in this disclosure is not limited to the above-mentioned loan screening decision prediction, and may be any prediction. In other words, the explanatory variables input to the prediction model and the objective variable output from the prediction model are not limited to the above-mentioned information.

[0012] The configuration and operation of a prediction model creation device 10 in this embodiment will be described below. The prediction model creation device 10 is configured with one or more information processing devices (computers) each including a calculation device and a storage device. As shown in FIG. 1, the prediction model creation device 10 includes a training data acquisition unit 11, a distribution estimation unit 12, a prediction model learning unit 13, and a prediction model output unit 14. The functions of the training data acquisition unit 11, the distribution estimation unit 12, the prediction model learning unit 13, and the prediction model output unit 14 can be realized by the calculation device executing a program for realizing each function stored in the storage device. The prediction model creation device 10 also includes a data storage unit 15 and a model storage unit 16. The data storage unit 15 and the model storage unit 16 are configured with a storage device.

[0013] The training data acquisition unit 11 (acquisition unit) acquires a set of explanatory variables and objective variables (x ij ,y ij ), and acquires training data consisting of averaged samples obtained by averaging a predetermined number of samples consisting of K raw data (step S1 in FIG. 2). Specifically, the training data acquisition unit 11 acquires training data D shown in Equation 1 by acquiring it from an external device or reading it from a storage medium, and stores it in the data storage unit 15. Here, the averaged samples shown in Equation 2 that make up the training data D are created by averaging a set of K raw data samples, as shown in Equation 3. In other words, the explanatory variables and the objective variable of the averaged samples are expressed by Equation 4.

number

number

number

number

[0014] Here, K is, for example, a hyperparameter that is set in advance and is common to the training data D. As an example, K may be 2 or another integer. Alternatively, K may be 1.5, and K is not limited to being an integer.

[0015] In this embodiment, the pair of explanatory variables and response variables (x ij ,y ij ) consists of explanatory variables such as each individual's age, annual income, savings, place of employment, and years of service before averaging, and a target variable such as whether or not they qualify for a loan. In other words, the raw data samples are actual examples of personal information. In contrast, averaged samples are data obtained by averaging multiple samples of personal information for each item, so individuals cannot be identified from such data. For example, if K=2 and the explanatory variable value of the averaged sample is 1, the raw data can have an infinite number of possible values, such as (1,1), (0,2), (-1,3), (-0.5,2.5), etc. Therefore, it is generally impossible to accurately restore each raw data sample from the averaged sample, and there is no risk of personal information being leaked from such averaged data.

[0016] As described above, the training data acquiring unit 11 is not limited to acquiring an averaged sample obtained by averaging a plurality of raw data samples from an external device, etc., but may acquire the raw data samples themselves from an external device, etc., and generate an averaged sample from the acquired plurality of samples. In this case, the training data acquiring unit 11 acquires the raw data samples and stores them in the data storage unit 15 in a way that prevents them from leaking to the outside.

[0017] The distribution estimation unit 12 (estimation unit) estimates a pre-averaging distribution P, which is a distribution of explanatory variables before averaging corresponding to explanatory variables constituting the averaged sample, which is the training data D acquired as described above (step S2 in FIG. 2). At this time, the distribution estimation unit 12 estimates the pre-averaging distribution P of the explanatory variables before averaging, which is shown in Expression 6, based on the averaged explanatory variables constituting the averaged sample shown in Expression 5.

number

number

[0018] Specifically, when the explanatory variables of the averaged sample shown in Equation 5 are given, the distribution estimation unit 12 utilizes domain knowledge regarding the distribution of explanatory variables to estimate the explanatory variables of the sample before averaging (x i1 ,...,x iK ) is the distribution of the prior average distribution. For example, we use Bayes' theorem to estimate the prior average distribution P as shown in Equation 7.

number

[0019] For example, if the averaged distribution of the explanatory variables of the averaging sample shown in the following formula 8 follows the standard normal distribution, then x ijThe distribution p of the explanatory variables of the sample before averaging can be estimated to follow a normal distribution with a mean of 0 and a standard deviation of 1 / √K (Equation 9 below), and can be estimated as shown in Equation 10. Note that in Equation 10, N(x; μ, σ 2 ) denotes the probability density of x in a normal distribution with mean μ and standard deviation σ.

number

number

number

[0020] Here, particularly when the averaging samples are averaged with K=2, the distribution P before averaging can be mathematically calculated as follows. In this case, if the post-averaging distribution of the explanatory variables of the averaging samples follows a normal distribution with a mean of 0 and a standard deviation τ, it can be calculated as in equation 11 and output as in equation 12. Note that δ is the Kronecker delta function.

number

number

[0021] As another example, if the explanatory variables of the averaged samples shown in the above formula 8 follow a uniform distribution, then x ij can be estimated assuming that the explanatory variables of the averaged samples shown in the above formula (8) follow a binomial distribution. ij can be estimated as following the Bernoulli distribution. Note that even in the case of distributions other than the normal distribution, it may be possible to calculate mathematically by using the convolution of the probability density function.

[0022] Here, since it may be difficult to accurately calculate the pre-averaging distribution p of the explanatory variables of the samples before averaging, the distribution estimation unit 12 may approximate the pre-averaging distribution by a sampling P, as will be described below. For example, the distribution estimation unit 12 generates the sampling P by using an importance sampling method through the following three steps. (1st step) p(x i1 ,...,x iK ) according to T random (x i1 ,...,x iK ) is generated as shown in Equation 13.

number

number

number

[0023] Note that other approximation methods for the distribution before averaging, such as the MCMC method, the Metropolis-Hastings algorithm, and Gibbs sampling, may also be used.

[0024] The prediction model learning unit 13 (learning unit) creates a prediction model f by machine learning a prediction model f that predicts a dependent variable from an explanatory variable using the averaged samples that are the training data D described above and the pre-averaging distribution P estimated as described above (step S3 in FIG. 2). At this time, the prediction model learning unit 13 first estimates the explanatory variables before averaging from the explanatory variables that constitute the averaged samples that are the training data D, based on the pre-averaging distribution P. Then, the prediction model learning unit 13 machine-learns the prediction model f so that the difference between the value based on the dependent variable predicted using the prediction model f from the estimated pre-averaging explanatory variables and the dependent variable that constitutes the averaged samples that are the training data D is small.

[0025] More specifically, the prediction model learning unit 13 performs machine learning of the prediction model f so as to minimize the negative log-likelihood L of the training data D, as shown in Equation 16 or Equation 17. As described above, when the pre-averaging distribution P is given, L in Equation 16 is minimized, and when sampling is given as an approximation of the pre-averaging distribution P, L in Equation 17 is minimized.

number

number

[0026] The function g in the above formulas 16 and 17 is the predictive variable x before averaging in the prediction model f shown in the following formula 18. ij The output when the input is ij It is a function that calculates the probability that the objective variable of the averaged sample shown in the following formula 19 will occur for the average of the two. The closer the two are to each other, the closer to 1 the value is output, and the further apart they are, the closer to 0 the value is output.

number

number

[0027] For example, the above function g can be expressed by Equation 20 in the case of regression, and by Equation 21 in the case of classification, where f(x) outputs the predicted probability. Note that β is a hyperparameter, and C is the number of classes.

number

number

[0028] And, minimizing the above L means to convert the averaged explanatory variables into the unaveraged explanatory variables x ij After estimating probabilistically, the explanatory variable x ij Output y for ij This is equivalent to training the prediction model f so that the average of and the averaged objective variable match.

[0029] As described above, when minimizing L, the upper limit of L may be calculated and then minimized. For example, the right-hand side of the following equation (22) may be minimized by using Jensen's inequality.

number

[0030] In this case, for example, in the case of regression, the right-hand side of the above equation (22) becomes the following equation (23), and therefore, by minimizing this right-hand side, the calculation becomes simpler.

number

[0031] As another example, the prediction model f may be approximated by a simpler function, and then L may be minimized. For example, the calculation of L may be made more efficient by Taylor expansion of the prediction model f(x) in the vicinity of the explanatory variables of the averaged sample shown in Equation 24.

number

number

[0032] The prediction model output unit 14 outputs the created prediction model f to a prediction device, which is another information processing device (step S4 in FIG. 2), or stores it in the model storage unit 16 of the prediction model creation device 10. The created prediction model is then used for prediction in the prediction device to which it was output or in the prediction model creation device 10. For example, by inputting into the created prediction model f an individual's age, annual income, savings, place of employment, years of service, etc. as explanatory variables necessary for an individual's loan screening, it is possible to obtain an output of a target variable such as whether or not to grant a loan. This can support decision-making by users of the prediction model f, such as those conducting loan screening.

[0033] As described above, in the present disclosure, when creating a prediction model f, machine learning is performed using training data obtained by averaging raw data such as personal information. Therefore, there is no risk of raw data such as personal information leaking from the training data itself or the prediction model, and security can be improved. Furthermore, when creating a prediction model f, machine learning is performed by estimating explanatory variables before averaging from averaged explanatory variables, so a highly accurate prediction model f can be created.

[0034] (Usage example) Next, a description will be given of an example of use of the present disclosure, taking as an example a case where a time until a sick patient is discharged from the hospital.

[0035] First, the prediction model creation device 10 acquires, as training data D, pairs of patient U's biometric information (explanatory variables) and the period until discharge (objective variable), as information on patient U who has already been discharged from the hospital. At this time, the prediction model creation device 10 acquires an averaged sample by averaging multiple pairs of biometric information and discharge period. Note that the biometric information of patient U may include, for example, patient U's age, sex, height, weight, occupation, blood type, medical history, genetic information, electronic medical record information, as well as blood pressure, heart rate, blood concentration, etc. measured using a wearable device or measuring device worn by the patient, as shown in FIG. 3.

[0036] Next, the predictive model creation device 10 estimates a pre-averaging distribution P, which is the distribution of explanatory variables before averaging that correspond to the explanatory variables that make up the averaged samples, which are the acquired training data D. At this time, the predictive model creation device 10 estimates the pre-averaging distribution, which is the distribution of explanatory variables before averaging, based on the post-averaging distribution, which is the distribution of explanatory variables of the averaged samples. Alternatively, the predictive model creation device 10 may approximate the pre-averaging distribution by sampling, for example, using the importance sampling method described above.

[0037] The prediction model creation device 10 then creates a prediction model f by machine learning a prediction model f that predicts a dependent variable from an explanatory variable using averaged samples that are training data D and the estimated pre-averaging distribution P. Specifically, the prediction model creation device 10 estimates the explanatory variables before averaging from the explanatory variables that constitute the averaged samples that are training data D, based on the pre-averaging distribution P. The prediction model creation device 10 then machine-learns the prediction model f so as to minimize the difference between the dependent variable predicted using the prediction model f from the estimated pre-averaging explanatory variables and the dependent variable that constitutes the averaged samples that are training data D. In this way, a prediction model f that predicts the length of hospital discharge from the biological information of patient U can be created.

[0038] Thereafter, by inputting the biometric information of a newly hospitalized patient U into the created prediction model f, it is possible to predict the time until discharge for that patient U. This can support the decision-making of medical professionals such as doctors.

[0039] As described above, by creating a prediction model f through machine learning using training data obtained by averaging raw data such as personal information, there is no risk of raw data such as personal information being leaked from the training data itself or the prediction model, thereby improving security. Furthermore, when creating the prediction model f, machine learning is performed by estimating the explanatory variables before averaging from the averaged explanatory variables, so that a highly accurate prediction model f can be created. Note that the above-described use example of the prediction model creation device 10 is merely an example, and the device may be used to create any prediction model.

[0040] <Second embodiment> Next, a second embodiment of the present disclosure will be described with reference to the drawings. In this embodiment, an overview of the predictive model creation device and the like described in the above-mentioned embodiment is shown. Note that the drawings may be relevant to any of the embodiments.

[0041] First, a description will be given of the hardware configuration of the predictive model creation device 100 according to the present disclosure. The predictive model creation device 100 is configured as a general information processing device, and is equipped with the following hardware configuration, as an example, as shown in FIG. ·CPU(Central Processing Unit)101(Arithmetic unit) ROM (Read Only Memory) 102 (storage device) RAM (Random Access Memory) 103 (storage device) Programs 104 loaded into RAM 103 A storage device 105 for storing a group of programs 104 A drive device 106 that reads and writes from a storage medium 110 external to the information processing device A communication interface 107 that connects to a communication network 111 outside the information processing device Input / output interface 108 for inputting and outputting data Bus 109 connecting each component

[0042] 4 shows an example of the hardware configuration of an information processing device that is the predictive model creation device 100, and the hardware configuration of the information processing device is not limited to the above-described case. For example, the information processing device may be configured with a part of the above-described configuration, such as not including the drive device 106. Furthermore, instead of the above-described CPU, the information processing device may use a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), an FPU (Floating point number Processing Unit), a PPU (Physics Processing Unit), a TPU (Tensor Processing Unit), a quantum processor, a microcontroller, or a combination thereof.

[0043] The CPU 101 acquires and executes the group of programs 104, thereby constructing and equipping the prediction model creation device 100 with the acquisition unit 121, estimation unit 122, and learning unit 123 shown in FIG. 5 . The group of programs 104 may be stored in advance in, for example, the storage device 105 or the ROM 102, and the CPU 101 loads the group of programs 104 into the RAM 103 and executes them as needed. The group of programs 104 may be supplied to the CPU 101 via the communication network 111, or may be stored in advance in the storage medium 110, with the drive device 106 reading out the programs and supplying them to the CPU 101. However, the acquisition unit 121, estimation unit 122, and learning unit 123 described above may be constructed using dedicated electronic circuits for realizing such means.

[0044] The acquisition unit 121 acquires training data consisting of averaged samples obtained by averaging a plurality of samples each consisting of a pair of an explanatory variable and a dependent variable. The estimation unit 122 estimates a pre-averaging distribution, which is a distribution of explanatory variables before averaging corresponding to the explanatory variables that constitute the averaged samples, which are the training data. The learning unit 123 performs machine learning to create a prediction model that predicts a dependent variable from an explanatory variable, based on the training data and the pre-averaging distribution.

[0045] With the above-described configuration, the present disclosure creates a predictive model through machine learning using training data obtained by averaging raw data such as personal information. This eliminates the risk of leakage of raw data such as personal information from the training data itself or the predictive model, thereby improving security. Furthermore, when creating a predictive model, machine learning is performed by estimating explanatory variables before averaging from averaged explanatory variables, making it possible to create a highly accurate predictive model.

[0046] At least one of the functions of the acquisition unit 121, estimation unit 122, and learning unit 123 described above may be executed by an information processing device installed and connected anywhere on the network, that is, may be executed by so-called cloud computing.

[0047] The above-described program can be stored and supplied to a computer using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / Ws, and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, and RAMs (Random Access Memory)). The program can also be supplied to a computer by various types of transitory computer-readable media. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer-readable media can supply the program to a computer via a wired communication path such as an electric wire or optical fiber, or via a wireless communication path.

[0048] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each of the above-described embodiments can be combined with other embodiments as appropriate.

[0049] <Additional Notes> Some or all of the above embodiments may be described as follows: The following provides an overview of the configurations of the predictive model creation device, predictive model creation method, and program according to the present disclosure. However, the present disclosure is not limited to the configurations described in the following supplementary notes. Note that the configurations described in Supplements 2 to 8, which are dependent on Supplementary Note 1 below, and some or all of the functions of the configurations, may also be dependent on other Supplements 9 and 10 in the same dependent relationship as Supplements 2 to 8. Furthermore, not limited to Supplements 1, 9, and 10, but also within the scope of the above-described embodiments, similar hardware, software, various recording means for recording software, or systems may be similarly made to be dependent on the configurations described as Supplements and some or all of the functions of the configurations. (Appendix 1) an acquisition unit that acquires training data consisting of averaged samples obtained by averaging a plurality of samples each consisting of a pair of an explanatory variable and a target variable; an estimation unit that estimates a pre-averaging distribution, which is a distribution of explanatory variables before averaging corresponding to explanatory variables that constitute the averaged samples that are the training data; a learning unit that performs machine learning to generate a prediction model that predicts a dependent variable from an explanatory variable based on the training data and the pre-averaging distribution; A predictive model creation device comprising: (Appendix 2) 10. The predictive model creation device according to claim 1, the learning unit performs machine learning to develop the prediction model using explanatory variables before averaging estimated from explanatory variables constituting the averaged samples, which are the training data, based on the pre-averaging distribution, and objective variables constituting the averaged samples, which are the training data. Predictive model creation device. (Appendix 3) 3. A predictive model creation device according to claim 2, the learning unit performs machine learning on the prediction model so as to reduce a difference between a value based on a dependent variable predicted using the prediction model from the estimated explanatory variables before averaging and a dependent variable constituting the averaged sample, which is the training data. Predictive model creation device. (Appendix 4) 3. A predictive model creation device according to claim 2, the learning unit performs machine learning on the prediction model so as to minimize a difference between an average value of a dependent variable predicted using the prediction model from the estimated explanatory variables before averaging and a dependent variable constituting the averaged sample, which is the training data. Predictive model creation device. (Appendix 5) 10. The predictive model creation device according to claim 1, the estimation unit estimates the pre-averaging distribution based on a post-averaging distribution which is a distribution of explanatory variables constituting the averaging samples which are the training data; Predictive model creation device. (Appendix 6) 10. The predictive model creation device according to claim 1, the estimation unit estimates the pre-averaging distribution such that explanatory variables before averaging of explanatory variables constituting the averaged samples, which are the training data, follow a normal distribution. Predictive model creation device. (Appendix 7) 10. The predictive model creation device according to claim 1, the estimation unit approximates and estimates the pre-averaging distribution by weighted sampling of explanatory variables before averaging of explanatory variables that constitute the averaged samples, which are the training data; Predictive model creation device. (Appendix 8) 10. The predictive model creation device according to claim 1, the acquisition unit acquires the training data consisting of averaged samples obtained by averaging two samples each consisting of a pair of an explanatory variable and a target variable. Predictive model creation device. (Appendix 9) Obtain training data consisting of averaged samples obtained by averaging a plurality of samples each consisting of a pair of explanatory variables and a target variable; Estimating a pre-averaging distribution, which is a distribution of explanatory variables before averaging corresponding to explanatory variables constituting the averaging sample, which is the training data; a prediction model that predicts a dependent variable from an explanatory variable based on the training data and the averaging prior distribution; Predictive modeling methods. (Appendix 10) 10. The method for creating a prediction model according to claim 9, further comprising: machine learning the prediction model using explanatory variables before averaging estimated from explanatory variables constituting the averaged samples, which are the training data, based on the distribution before averaging, and objective variables constituting the averaged samples, which are the training data; Predictive modeling methods. (Appendix 11) 11. The method of creating a predictive model according to claim 10, further comprising: machine learning the prediction model so that the difference between a value based on a dependent variable predicted using the prediction model from the estimated explanatory variables before averaging and a dependent variable constituting the averaged sample, which is the training data, is small; Predictive modeling methods. (Appendix 12) 11. The method of creating a predictive model according to claim 10, further comprising: machine learning the prediction model so as to minimize the difference between an average value of a dependent variable predicted using the prediction model from the estimated explanatory variables before averaging and a dependent variable constituting the averaged sample, which is the training data; Predictive modeling methods. (Appendix 13) 10. The method for creating a prediction model according to claim 9, further comprising: estimating the pre-averaging distribution based on a post-averaging distribution which is a distribution of explanatory variables constituting the averaging samples which are the training data; Predictive modeling methods. (Appendix 14) 10. The method for creating a prediction model according to claim 9, further comprising: estimating the pre-averaging distribution so that explanatory variables before averaging constituting the averaging sample, which is the training data, follow a normal distribution; Predictive modeling methods. (Appendix 15) 10. The method for creating a prediction model according to claim 9, further comprising: approximating and estimating the pre-averaging distribution by weighted sampling of explanatory variables before averaging of explanatory variables constituting the averaged samples, which are the training data; Predictive modeling methods. (Appendix 16) 10. The method for creating a prediction model according to claim 9, further comprising: Acquire the training data consisting of averaged samples obtained by averaging two samples each consisting of a pair of an explanatory variable and a target variable. Predictive modeling methods. (Appendix 17) Obtain training data consisting of averaged samples obtained by averaging a plurality of samples each consisting of a pair of explanatory variables and a target variable; Estimating a pre-averaging distribution, which is a distribution of explanatory variables before averaging corresponding to explanatory variables constituting the averaging sample, which is the training data; a prediction model that predicts a dependent variable from an explanatory variable based on the training data and the averaging prior distribution; A program that causes a computer to perform a process. [Explanation of symbols]

[0050] 10. Prediction model creation device 11 Training data acquisition section 12 Distribution estimation part 13 Prediction model training section 14 Prediction model output section 15 Data storage unit 16 Model memory section 100 Prediction model creation device 101 CPU 102 ROM 103 RAM 104 Programs 105 Storage device 106 Drive device 107 Communication Interface 108 Input / Output Interface 109 Bus 110 Storage medium 111 Communication Network 121 Acquisition Department 122 Estimation Department 123 Learning Department

Claims

1. an acquisition unit that acquires training data consisting of averaged samples obtained by averaging a plurality of samples each consisting of a pair of an explanatory variable and a target variable; an estimation unit that estimates a pre-averaging distribution, which is a distribution of explanatory variables before averaging corresponding to explanatory variables that constitute the averaged samples that are the training data; a learning unit that performs machine learning to generate a prediction model that predicts a dependent variable from an explanatory variable based on the training data and the pre-averaging distribution; A predictive model creation device comprising:

2. The predictive model creation device according to claim 1, the learning unit performs machine learning to develop the prediction model using explanatory variables before averaging estimated from explanatory variables constituting the averaged samples, which are the training data, based on the pre-averaging distribution, and objective variables constituting the averaged samples, which are the training data. Predictive model creation device.

3. 3. The predictive model creation device according to claim 2, the learning unit performs machine learning on the prediction model so as to reduce a difference between a value based on a dependent variable predicted using the prediction model from the estimated explanatory variables before averaging and a dependent variable constituting the averaged sample, which is the training data. Predictive model creation device.

4. 3. The predictive model creation device according to claim 2, the learning unit performs machine learning on the prediction model so as to minimize a difference between an average value of a dependent variable predicted using the prediction model from the estimated explanatory variables before averaging and a dependent variable constituting the averaged sample, which is the training data. Predictive model creation device.

5. The predictive model creation device according to claim 1, the estimation unit estimates the pre-averaging distribution based on a post-averaging distribution which is a distribution of explanatory variables constituting the averaging samples which are the training data; Predictive model creation device.

6. The predictive model creation device according to claim 1, the estimation unit estimates the pre-averaging distribution such that explanatory variables before averaging of explanatory variables constituting the averaged samples, which are the training data, follow a normal distribution. Predictive model creation device.

7. The predictive model creation device according to claim 1, the estimation unit approximates and estimates the pre-averaging distribution by weighted sampling of explanatory variables before averaging of explanatory variables that constitute the averaged samples, which are the training data; Predictive model creation device.

8. The predictive model creation device according to claim 1, the acquisition unit acquires the training data consisting of averaged samples obtained by averaging two samples each consisting of a pair of an explanatory variable and a target variable; Predictive model creation device.

9. Obtain training data consisting of averaged samples obtained by averaging a plurality of samples each consisting of a pair of explanatory variables and a target variable; Estimating a pre-averaging distribution, which is a distribution of explanatory variables before averaging corresponding to explanatory variables constituting the averaging sample, which is the training data; a prediction model that predicts a dependent variable from an explanatory variable based on the training data and the averaging prior distribution; Predictive modeling methods.

10. Obtain training data consisting of averaged samples obtained by averaging a plurality of samples each consisting of a pair of explanatory variables and a target variable; Estimating a pre-averaging distribution, which is a distribution of explanatory variables before averaging corresponding to explanatory variables constituting the averaging sample, which is the training data; a prediction model that predicts a dependent variable from an explanatory variable based on the training data and the averaging prior distribution; A program that causes a computer to perform a process.

Images