JP6567484B2 - Estimated model construction system, estimated model construction method and program - Google Patents

Description

  The present invention relates to an estimation model construction technique.

  Insurers such as local governments and health insurance unions need to implement effective and efficient medical measures in order to achieve both health promotion and health cost control for residents and workers who are insured. As one of the means for realizing this, the utilization of electronic information accumulated in the past is attracting attention. For example, using accumulated information such as basic resident information and medical fee details, quantifying the characteristics of each group such as regions and offices, for example, age distribution, insurance use status, number of sick people, etc. Appropriate medical measures according to the characteristics can be selected and applied.

  Furthermore, in addition to the current characteristics, it is expected that more appropriate medical measures can be selected by estimating future situations through data analysis. For example, Patent Document 1 proposes a patient share analysis technique of a medical institution that predicts the future population and the number of patients by age and disease based on mesh population and patient incidence data by international disease classification.

Japanese Patent Laid-Open No. 2006-111986

  When estimating the future using stored information, not only a fixed specific group but also several different granularities, such as prefectures, cities, school districts, etc., can be freely selected as the target of estimation. If both can be estimated with high accuracy, it can be expected that more detailed medical measures can be implemented. However, insurers are a mixture of people with various backgrounds (attributes and past histories) that vary greatly in health status, visit history, etc., so it is difficult to estimate with high accuracy if the target group size is small was. Patent Document 1 is an estimation of only a fixed group, and the granularity (group attribute, etc.) cannot be changed.

  Therefore, the present invention has been made in view of the above problems, and an object thereof is to perform future estimation with high accuracy even in a group in which people with different backgrounds are mixed.

  The present invention is an estimation model construction system that includes a processor that executes a program and a memory that stores the program, and constructs an estimation model by executing the program, and is an accumulation related to medical and health for each individual. A storage unit that stores information, a data item included in the accumulated information, and related category information that stores a correspondence relationship between a data item that is another data item related to the data item and includes category information; The first objective variable that is the estimation target of the estimation model and the explanatory variable that is the input information of the estimation model are determined based on the received information or the accumulated information, and based on the correspondence relationship of the related category information , Determining a categorical variable corresponding to the first objective variable, and the stored information in which a value of the categorical variable satisfies a predetermined condition An estimation model constructing unit that divides and constructs a first estimation model corresponding to each of the divided accumulated information, the first objective variable, the explanatory variable, and the categorical variable as information to be estimated, and the estimation A future estimation unit that selects the first estimation model to be applied according to the value of the categorical variable included in the target information.

  According to the representative embodiment of the present invention, it is possible to construct an estimation model capable of highly accurate future estimation even in a group in which people with different backgrounds are mixed.

  Problems, configurations, and effects other than those described above will become apparent from the description of the following embodiments.

1 is a block diagram illustrating a configuration of an estimation model construction system according to a first embodiment of this invention. FIG. It is a figure which shows Example 1 of this invention and shows an example of shaping data. It is a flowchart which shows Example 1 of this invention and shows an example of an estimation model construction process. 5 is a flowchart illustrating an example of related category reading processing according to the first embodiment of this invention. It is a figure which shows Example 1 of this invention and shows an example of a related category matrix. It is a tree which shows Example 1 of this invention and shows an example of the relationship between related categories. It is a tree which shows Example 1 of this invention and shows the other example of the relationship between related categories. It is a figure which shows Example 1 of this invention and shows an example of the shaping data for estimation model construction. It is a figure which shows Example 1 of this invention and shows an example of a labeling table. It is a figure which shows Example 1 of this invention and shows an example of a labeling table. It is a figure which shows Example 1 of this invention and shows the other example of the shaping data for estimation model construction. It is a figure which shows Example 1 of this invention and shows the other example of a labeling table. It is a figure which shows Example 1 of this invention and shows the other example of a labeling table. It is a figure which shows Example 1 of this invention and shows the other example of a labeling table. It is a figure which shows Example 1 of this invention and shows the other example of a labeling table. It is Example 1 of this invention, and is a flowchart which shows an example of an estimation process. It is a figure which shows Example 1 of this invention and shows an example of an operation screen. It is a figure which shows Example 1 of this invention and shows an example of an estimation result display screen. It is Example 1 of this invention, and is a flowchart which shows an example of an estimation information aggregation process.

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

  In the first embodiment, when the estimation model is constructed from the accumulated information collected in advance, regarding the categorical variable strongly related to the target variable to be estimated, the data is divided for each record having the same value, and each value is divided. An example of an estimation model construction system that constructs an estimation model corresponding to the above will be described.

  Hereinafter, the configuration of the estimation model construction system of the first embodiment will be described. FIG. 1 is a block diagram illustrating the configuration of the estimation model construction system according to the first embodiment. The estimation model construction system according to the first embodiment includes a data analysis unit 101 and a database 120. The data analysis unit 101 includes an input unit 102, an output unit 103, an arithmetic device 104, a memory 105, and a storage medium 106.

  The input unit 102 is a human interface such as a mouse or a keyboard, and accepts input to the data analysis unit 101. The output unit 103 is a display or a printer that outputs a calculation result obtained by the estimation model construction system. The storage medium 106 is a storage device that stores various programs for realizing data analysis processing by the data analysis unit 101, execution results of the data analysis processing, and the like. For example, a non-volatile storage medium (magnetic disk drive, non-volatile memory, etc.) ).

  In the memory 105, a program stored in the storage medium 106 is expanded. The arithmetic device 104 is an arithmetic device that executes a program loaded in the memory 105, and is, for example, a CPU, a GPU, or the like. The processing device 104 executes the processing and calculation described below.

  The estimation model construction system of the first embodiment is a computer system that is configured on a single computer or on a plurality of computers that are logically or physically configured, and in a separate thread on the same computer. It may operate, and may operate on a virtual machine constructed on a plurality of physical computer resources.

  A program executed by the arithmetic device 104 is provided to each server via a removable medium (CD-ROM, flash memory, etc.) or a network, and is stored in a nonvolatile storage device that is a non-temporary storage medium. For this reason, the computer system may include an interface for reading removable media.

  The data analysis unit 101 causes the data shaping unit 107 stored in the storage medium 106, the estimation model construction unit 108, the future estimation unit 111, the estimation information aggregation unit 112, and the related category association unit 113 to function.

  The arithmetic unit 104 operates as a functional unit that provides a predetermined function by performing processing according to a program of each functional unit. For example, the arithmetic device 104 functions as the data shaping unit 107 by performing processing according to the data shaping program. The same applies to other programs. Furthermore, the arithmetic unit 104 also operates as a function unit that provides the functions of a plurality of processes executed by each program. A computer and a computer system are an apparatus and a system including these functional units.

  Hereinafter, the types of data handled in the first embodiment and the estimation model to be constructed will be described. In the first embodiment, an example will be described in which data including receipt information and specific medical examination information for each individual in time series is used as data.

  The estimation target of the estimation model to be constructed is the estimation of the risk of medical institution visits and the expected value of medical expenses for each individual regarding future diseases, and the number of medical institution visits and the medical expenses of the group that statistically aggregates them. An example will be described.

  First, the data shaping unit 107 will be described. The data shaping unit 107 stores information on medical care and health for each individual such as the receipt information stored in the data storage unit 121 and the medical remuneration statement for each individual of the specific medical examination information for each specific period and for each individual. The aggregated shaping data 201 for estimation is constructed and stored in the shaping data storage unit 122. In the first embodiment, the shaping data 201 is handled as accumulated information related to medical care and health for each individual.

  Information stored in each of the storage units 121 to 128 is not limited to the database 120, and may be stored in a storage unit or a storage device accessible from the data analysis unit 101.

  FIG. 2 is a diagram for explaining an example of the shaping data for estimation shaped by the data shaping unit 107. In the first embodiment, as an example, an example will be described in which medical information such as receipt information for each individual is aggregated every year.

  Information 201 represents the entire shaped data. Hereinafter, information included in the shaping data 201 will be described. The basic information 210 is information including basic profile information for each individual.

  Here, as an example, an example including a personal ID 211, a gender 212, and an age 213 is shown. The personal ID 211 is a personal ID uniquely assigned to each individual included in the shaping data 201. The sex 212 is information representing the sex for each individual. For example, males can be represented as categorical variables assigned to specific variables, such as 0, females 1, and so on. Age 213 is information representing the age of each individual. For example, it can be expressed as an integer such as 35, 21, 50.

  In the figure, reference numeral 220 denotes year information including information for each individual aggregated for each year. Here, the example which gathered the information of 2014 is shown. The specific medical examination information 221 includes information related to the specific medical examination received in the current year, and includes information such as BMI (Body Mass Index), blood sugar, blood pressure, and the like. In the figure, information of BMI 223 is displayed as an example.

  The BMI 223 includes information on the BMI measured in 2014. The information included here may include information on an inquiry in a specific medical examination, for example, questionnaire information regarding exercise habits, questionnaire information regarding drinking, smoking, and the like. In this case, as the data, for example, the data is stored as a categorical variable including a symbol or a numerical number assigned for each amount of drinking per week.

  The receipt information 222 includes information regarding a receipt generated in the current fiscal year. For example, information such as the presence / absence of a receipt for each wound name, the number of occurrences of the receipt, and related medical costs, which are defined based on the wound name included in the receipt, is included. In addition, information on each disease and the like in which the names of wounds are grouped into diseases based on known aggregated information are included. In the figure, information about diabetes, diabetes medical expenses, and hypertension is shown as an example.

  Diabetes mellitus 224 is information indicating the occurrence of a diabetes-related receptor. In the illustrated example, 1 indicates that at least one receptor related to diabetes has occurred in the current year, and 0 indicates that it has not occurred. The diabetes medical cost 225 is information indicating a medical cost included in a receipt regarding diabetes. In the example shown in the figure, information on medical expenses incurred in connection with diabetes in the current year is included.

  The hypertension 226 is information indicating the occurrence of a receptor related to hypertension, and in the illustrated example, 1 indicates that at least one receptor related to hypertension has occurred in the corresponding year, and 0 indicates that it has not occurred. Yes.

  In the figure, reference numeral 230 denotes year information including information for each individual aggregated for each year. The example shown in the figure shows an example in which the information for 2015 is collected, but the counting method is the same as the year information 220 for 2014. The BMI 231 is information including information on the BMI acquired in the specific medical examination in 2015, and is information stored in the table in the same manner as the information in 2014. By referring to the shaping data 201 described above, it is possible to acquire basic information for each individual and information on health and medical care for each year at a time.

  Next, the estimation model construction unit 108 will be described. The estimation model construction unit 108 executes an estimation model construction process, and constructs an estimation model based on the shaped data 201 stored in the shaped data storage unit 122.

  Hereinafter, an example of the estimation model construction process processed by the estimation model construction unit 108 will be described. In the detailed description of the process, first, an outline of the estimation model constructed by this process will be described. Next, it will be described that two different forms of estimation models can be constructed according to the characteristics of information used for model construction when the estimation model is constructed. Finally, how each of the two different types of estimation models is constructed will be described step by step using the flowchart of the processing shown in FIG.

  First, the outline of the estimation model will be described. The estimation model constructed in the first embodiment is constructed from the shaped data 201. For example, one objective variable, one or more explanatory variables, and a relationship between variables for estimating the objective variable from the explanatory variables. This is an estimation model that includes information on sex.

  As an estimation model, for example, the presence or absence of a medical institution related to diabetes in 2015 based on receipt information (takes a value of 0 or 1) as an objective variable, and the BMI in 2014 based on a specific medical checkup (takes a continuous value) 223 and an estimation model having fasting blood glucose (taking a continuous value) as explanatory variables can be constructed.

  This estimation model can estimate the probability of whether or not to see a medical institution related to diabetes in 2015 based on the BMI 223 and fasting blood glucose value in 2014 using the relationship between variables learned in advance.

  In this way, it is possible to estimate future information by constructing an estimation model in which information having different acquisition times is set as an explanatory variable and an objective variable. In the case of this estimation model, by inputting information on BMI and fasting blood glucose in a certain year, it is possible to estimate the probability of having a medical institution with diabetes in the next year.

  For example, by inputting BMI information and fasting blood glucose information of an individual in this estimation model in 2015, the value of whether or not the individual has received a medical institution in 2016 can be estimated. For example, when the values of presence / absence of consultation are 0 and 1, the probability value of presence / absence of consultation can be estimated as a value of 0 or more and 1 or less.

  Next, different types of estimation models that can be constructed by the estimation model construction unit 108 in the first embodiment will be described. In the first embodiment, the estimation model construction unit 108 divides data for each value of the categorical variable among the categorical variables included in the explanatory variables in the estimation model, and sub-models are obtained from each data. Build up. Furthermore, based on the type of this categorical variable, two different forms of estimation models are constructed.

  Specifically, when the categorical variable applies the estimation model, that is, when the information of the explanatory variable necessary for the estimation model to estimate the information for the target year to be estimated is input, Two different forms of estimation models can be constructed according to two conditions: (1) information that has been confirmed and (2) information that has not been confirmed. . Hereinafter, two forms of the estimation model will be described.

  The estimation model of the first form is an estimation model in a case where information included in the categorical variable C in the explanatory variable is information that has been confirmed when the estimation model is applied. For example, as an estimation model, BMI 223 in 2014, fasting blood glucose in 2014, and the presence or absence of diabetes 224 in 2014 are explanatory variables, and the presence or absence of diabetes in 2015 is an objective variable. An estimation model is an example.

  Here, it is assumed that the estimation model construction unit 108 selects the value of the presence or absence of diabetes 224 in 2014 included in the explanatory variable as a categorical variable for dividing the data and constructing a submodel. Since this estimation model is constructed as a model for estimating the information for the year 2015 from the information for the year 2014, the information for the year N + 1 is estimated when the information for the year N is input.

  At this time, the value of the presence / absence of diabetes in 2014 designated as the categorical variable is the information of 2014 as well as the information on the 2014 BMI and fasting blood glucose as other explanatory variables. is there. For example, when estimating the information of FY2016 from the information of FY2015 using this estimation model, it is estimated by inputting the information of FY2015 that has been confirmed to the value of the presence or absence of diabetes in FY2014 that is a categorical variable. it can. As described above, when the information included in the categorical variable C of the explanatory variable is information that has been confirmed when the estimation model is applied, the value of the categorical variable is the same (or a predetermined value) as the estimation model of the first form. The shaping data 201 is divided for each piece of data satisfying the conditions, and an estimation model corresponding to each categorical variable value is constructed from each shaping data 201. Note that the values of the categorical variables are not necessarily the same, and depending on the type of the shaped data 201, the shaped data 201 may be divided for each piece of data that satisfies a predetermined condition such as an approximate value (difference is less than a threshold). Good.

  The estimation model of the second form is an estimation model in a case where information included in the categorical variable C in the explanatory variables is uncertain information when the estimation model is applied. For example, as an estimation model, an estimation model in which the presence or absence of a disease A visit in 2014 and the presence or absence of a disease A in 2015 is an explanatory variable and the presence or absence of a disease B visit in 2015 is an example variable.

  Here, it is assumed that the estimation model construction unit 108 selects whether or not to receive the disease A in 2015 among the explanatory variables as a categorical variable for dividing the data and constructing a submodel. Since this estimation model is built as a model that estimates the information of other 2015 from the information of 2014 and 2015, other N + 1 years based on the input of N year information and N + 1 year information Is a model for estimating

  At this time, the information on whether or not to receive the disease A in 2015 designated as the categorical variable is information acquired in the same year as the information on whether or not to receive the disease B in 2015 as the objective variable. Therefore, when this estimation model is applied, it is considered that the information of this categorical variable is unconfirmed.

  For example, when using this estimation model to estimate information for FY2016 from information for FY2015, the information on disease A in FY2016, which is a categorical variable, is different from other explanatory variables, and information is not available at the time of estimation. It is definite. In this case, first, a first estimation model for estimating the value of the categorical variable is constructed based on the information of the explanatory variables other than the categorical variable. Next, the shaped data is divided for each data having the same value of the categorical variable, and a second estimation model corresponding to the value of the categorical variable is constructed from each data.

  For example, in the above-described example, as the first estimation model, the first estimation model is constructed in which the information on whether or not to receive the disease A in 2014 is an explanatory variable, and the information on whether or not the disease A is received in 2015 is the objective variable. To do. Next, as a second estimation model, an estimation model in which information on whether or not to see a disease A in 2014 is used as an explanatory variable, and information on whether or not a disease B is received in 2015 is an objective variable. Build for each data set with the same value of whether or not there is a medical examination.

  When estimating the information of FY2016 from the information of 2015 using this estimation model, first, using the first estimation model, from the information on whether or not the disease A was received in 2015, the disease A of 2016 Estimate the probability distribution of the presence or absence of a visit. Next, the second estimation model is applied under the estimated probability distribution of the presence or absence of the disease A in 2016, and the results are integrated to obtain the final estimation result.

  These two different types of estimation models are constructed by selecting any appropriate one based on the information included in the shaping data 201 and the information stored in the database 120.

  Hereinafter, a flowchart of the estimation model construction process performed by the estimation model construction unit 108 will be described with reference to FIG.

  FIG. 3 is a flowchart for explaining the flow of the estimation model construction process executed by the estimation model construction unit 108. Hereinafter, the processing of each step will be described using this flowchart.

  In the objective variable / explanatory variable input step 301, the estimation model construction unit 108 determines a variable to be estimated as an objective variable based on information input from the input unit 102, and determines information used for estimation as an explanatory variable. To do. The correspondence between the explanatory variable and the objective variable is determined by storing the information of the explanatory variable and the objective variable as a pair in advance in the estimation model construction variable group storage unit 127 and reading out the previously stored information. You may decide by.

  In step 302 of the shaping data reading process, the estimation model construction unit 108 stores the information of the shaping data 201 for construction of the estimation model previously processed by the data shaping unit 107 and stored in the shaping data storage unit 122. Read from the unit 122.

  In step 303 of the related categorical variable reading process, the estimation model construction unit 108 includes a variable that should be treated as a categorical variable in estimating the objective variable among the variables included in the explanatory variable based on the objective variable information. It is determined whether or not. This determination is performed, for example, when the estimation model construction unit 108 uses information of the related category matrix (related category information) 501 stored in the related category storage unit 124.

  FIG. 5 shows matrix information for selecting a related category stored in the related category storage unit 124. The related category matrix 501 is a matrix that sets information on target variables that are estimation targets in the row direction, and sets information on related variables in the column direction. Each row and column includes items included in the shaped data 201. Information is included.

  The target variable information 502 in the row direction includes information related to the target variable from which the association is searched. The column-related related variable information 503 includes information on items to be searched for relationships.

  In the example shown in the figure, as the basic information B and N year information, A, y1, y2, and N + 1 year information including A, y1, and y2 are included. The matrix contains 0, 1, or empty information. An empty cell indicates that the target variable of the row including the cell does not have an association with the related variable of the column including the cell.

  For a cell containing 0, the target variable of the row containing the cell has an association with the related variable of the column containing the cell, and the related variable has already been observed at the time of estimation. Information, that is, information at the time of estimation indicates that the information is included in the shaped data.

  For cells that contain 1, the target variable in the row that contains the cell has an association with the variable in the column that contains the cell, and information that has not been observed at the time of estimation, that is, information that has been formatted at the time of estimation. Indicates that it is not included in the data. For example, if the related variable is a variable related to information in the past year than the target variable, the variable is information that has been observed. On the other hand, if the related variable is information acquired in the same year as the target variable, the variable is unobserved information at the time of estimation.

  A process of reading a related category using the related category matrix 501 will be described with reference to the flowchart of FIG. FIG. 4 is a flowchart of the process executed in step 303 of the related category variable reading process of the estimation model construction process.

  In step 401 for determining the target variable, the estimation model construction unit 108 determines a variable for searching for a relationship. Specifically, the estimation model construction unit 108 determines the objective variable to be estimated, which is determined in step 301 for inputting the objective variable and explanatory variable of the estimation model construction processing, as the original variable to be searched.

  In step 402 of the related category variable search, the estimation model construction unit 108 searches for a category variable related to the target variable using the related category matrix 501. Specifically, the estimation model construction unit 108 searches the related category matrix 501 for a row that includes a variable that is a target variable, and searches for a cell that includes 0 or 1. When there is a cell including 0 or 1, the estimation model construction unit 108 stores the related variable of the corresponding column in association with the value in the cell as the related categorical variable corresponding to the target variable. .

  The related category matrix shown in FIG. 5 will be described as an example. For example, when the target variable is information on A in N + 1 year, the estimation model construction unit 108 selects A in N year, which is a column including a cell in which 0 exists, as a related variable. For example, when the target variable is information on y1 in N + 1 year, the estimation model construction unit 108 selects information on A in N + 1 year that is a column including a cell in which 1 exists as a related categorical variable.

  In step 403, the estimation model construction unit 108 performs a determination on the related category variable stored in step 402 of variable search. The estimation model construction unit 108 determines whether or not the selected related category variable includes an undetermined categorical variable at the time of estimation. If included, the process proceeds to step 404 for reading the undetermined category item. On the other hand, the estimation model construction unit 108 proceeds to step 405 of the related variable information aggregation process when the categorical variable that is not determined at the time of estimation is not included or no related category is stored.

  The estimation model construction unit 108 determines whether or not the categorical variable is undetermined at the time of estimation based on whether the value of the cell corresponding to the related category variable selected from the related category matrix is 0 or 1. It can be determined based on.

  In step 404 of making the uncertain category item target variable, the estimation model construction unit 108 sets the uncertain category variable at the time of estimation among the related variables stored in step 402 as a new target variable, and then the related category variable. Return to step 402 of the search. Thereafter, the above-described steps 402, 403, and 404 are repeated repeatedly until the processing of all undetermined related category variables is completed.

  The determination in step 403 and the repetition processing in steps 402, 403, and 404 will be described using the related category matrix 501 shown in FIG. 5 as an example.

  For example, when the target variable is A in N + 1 year, the estimation model construction unit 108 stores A in N year, which is a column including a cell in which 0 exists, as a related category variable. In the related category matrix 501, 0 indicates that the related variable is a confirmed variable in the related relationship with the target variable. Therefore, the estimation model construction unit 108 determines No in the determination in step 403 and proceeds to step 405.

  On the other hand, when the target variable is y1 in the year N + 1, the estimation model construction unit 108 selects A in the year N + 1, which is a column including the cell in which 1 exists, as the related category variable. In the related category matrix 501, 1 indicates that the related variable is an undetermined variable in the related relationship with the target variable. Therefore, the estimation model construction unit 108 determines Yes in the determination in step 403, and proceeds to step 404.

  In step 404, the estimation model construction unit 108 sets A for N + 1 year as a new target variable. Thereafter, in step 402 again, the estimation model construction unit 108 stores A in N year as a categorical variable related to A in N + 1 year. Since this variable is a variable that includes a value of 0 in the cell and is fixed at the time of estimation in relation to A in the year N + 1, the estimation model construction unit 108 determines No in the second step 404 and repeats. The process ends, and the process proceeds to step 405.

  In step 405 of related variable information aggregation, the estimation model construction unit 108 aggregates information on the related categorical variables stored in the processes so far, and constructs a tree structure indicating the relevance of the related categorical variables.

  Hereinafter, the tree structure constructed in this process will be described with reference to FIG. 6A and 6B are tree structures representing examples of the relevance of variables read out in the related category variable reading process using the related category matrix 501 in FIG.

  FIG. 6A shows a tree structure when the target variable is set to A in the year N + 1 and the related category variable reading process is performed. In this example, one A in N year is read as a categorical variable related to the target variable. FIG. 6A shows a tree structure representing a relationship, in which a square represents a variable and an arrow represents a relationship.

  In FIG. 6A, A_N + 1 represents a variable representing A information in year N + 1, and A_N represents a variable representing A information in year N. In the arrow, the variable at the root is the related variable, and the variable at the tip of the arrow is the target variable affected by the related variable at the root.

  In this way, when the related categorical variable related to the target variable is a variable that has been determined when the objective variable is estimated, the distance on the tree structure between the target variable and the related categorical variable is 1. This tree structure corresponds to the estimation model of the first form.

  FIG. 6B shows a tree structure when the target variable is set to the information of y1 in the year N + 1 and the related category variable is read. In this example, as the categorical variable related to the target variable, A in N + 1 year is read as the first related variable, and then the information of A in N + 1 year is uncertain at the time of estimation. Is read as a second related variable. At this time, there is a relationship in which the information of y1 in the year N + 1 is related to the information of A in the year N + 1 and the information of A in the year N + 1 is related to the information of A in the year N.

  FIG. 6B is a tree structure showing the above-mentioned relevance information, where y1_N + 1 is a variable meaning y1 information in N + 1 year, A_N + 1 is a variable meaning A information in N + 1 year, and A_N is in N year A variable representing the information of A is represented. FIG. 6B shows a relationship in which N_N affects A_N + 1, and A_N + 1 affects y1_N + 1.

  As described above, when the related categorical variable related to the target variable includes a variable that is not determined at the time of estimation, the depth of the tree structure of the target variable and the related categorical variable is 2 or more. This tree structure corresponds to the estimation model of the second form. In the example of FIG. 6B, y1_N + 1 indicates the root node, and A_N + 1 indicates the internal node of the related categorical variable.

  The estimation model construction unit 108 stores the tree structure constructed by this processing in the related category storage unit 124.

  This is the end of the description of step 303 of the related category variable reading process in the estimation model construction process. Hereinafter, the description returns to the estimation model construction process using FIG.

  In step 304 of the data division labeling process, the estimation model construction unit 108 performs a labeling process for classifying individual data to perform data division performed at the time of construction of the estimation model.

  Specifically, first, the estimation model construction unit 108 selects related category variable information read in step 303 of the related category variable reading process from the explanatory variables included in the shaping data 201. Next, the estimation model construction unit 108 constructs an estimation model for estimating the relevant variable for each target variable based on the tree structure between the relevant category variables constructed in Step 303 of the relevant category variable reading process. Perform labeling for data partitioning. Below, each example is demonstrated using two tree structure of FIG. 6A and FIG. 6B.

  First, an example of processing the tree structure shown in FIG. 6A will be described.

  In the tree structure of FIG. 6A, the objective variable is A_N + 1 and the categorical variable is A_N. FIG. 7 is a diagram illustrating an example of the shaping data 2200 for estimating model construction. Estimated model construction shaping data 2200 indicates the result of selecting information on the item of the related category variable from the shaping data 201. The estimated model construction shaping data 2200 includes No 2201, a confirmed category 2202, an objective variable 2203, and explanatory variables 2204 to 2205 in one entry.

  In the first embodiment, it is assumed that the categorical variable A_N and the objective variable A_N + 1 are variables that can take two types of values, y and n. For example, y is set when corresponding to a category, and n is set when not included in a category, for example. In step 304 of the data division labeling process, labeling for data division is performed for each variable that needs to be estimated using the information of the shaping data 2200 for constructing the estimation model.

  In the tree structure of FIG. 6A, the estimation model construction unit 108 labels the data for the data division processing for constructing the model for estimating A_N + 1 because only one of the objective variables A_N + 1 is required to be estimated. Do. This label is set so that all combinations of states of variable groups that are ancestors of variables to be estimated are unique in the tree structure.

  For example, it can be implemented by labeling with a variable that takes the direct product value of the ancestor variable. In the tree of FIG. 6A, since there is only one A_N ancestor node of A_N + 1, it can be labeled by directly using the value of A_N.

  FIG. 8A is a labeling table 2300 showing the result of assigning labels for data division with respect to the tree of FIG. 6A. The related category 2301 is a variable used for labeling. The data division label 2302 is a given label. In the illustrated example, the value of A_N is used as it is for the label.

  FIG. 8B shows an example of the labeling table 2310. The labeling table 2310 includes No 2311, a confirmed category 2312, an estimation target 2313, and a data division label 2314 in one entry.

  FIG. 8B is an example in which the estimation model construction shaping data 2200 shown in FIG. 7 is labeled using the label shown in FIG. 8A. The data division label 2314 is a column of labels generated based on categorical variable information.

  Next, an example of processing the tree shown in FIG. 6B will be described.

  FIG. 9 is a diagram illustrating an example of the shaping data 2210 for estimating model construction. Estimated model construction shaping data 2210 indicates a result of selecting related category variable item information from shaping data 201 based on the information of FIG. 6B. The estimated model construction shaping data 2210 includes No 2211, a confirmed category 2212, an unconfirmed category 2213, an objective variable 2214, and explanatory variables 2215 to 2216 in one entry.

  In the example of FIG. 6B, the objective variable is y1_N + 1, and there are two categorical variables, A_N + 1 and A_N. Of these, A_N is a confirmed category, and A_N + 1 is an unconfirmed category.

  Here, it is assumed that the categorical variables A_N and A_N + 1 are variables that can take two types of values, y and n. In step 304 of the data division labeling process, based on this information, for each variable that requires estimation, labeling is performed when data is divided when an estimation model for estimating the variable is constructed.

  The variables that need to be estimated in the tree of FIG. 6B are the target variable y1_N + 1 and the categorical variable A_N + 1. Therefore, two labeling processes are executed: labeling in the data division process for constructing the estimation model for estimating y1_N + 1 and labeling in the data division process for constructing the model for executing the estimation of A_N + 1.

  Similar to the example of FIG. 6A, these labeling can also be performed by labeling so that combinations of values of variables that are ancestors of variables to be estimated are unique.

  FIG. 10A is a diagram illustrating an example of the labeling table 2320. FIG. 10A is a labeling table 2320 showing a result of giving a data division label for model construction for estimating the categorical variable A_N + 1 with respect to the tree of FIG. 6B.

  The related category 2321 is a variable used for labeling. The data division label 2322 is a given label. In this example, since the variable serving as the ancestor of A_N + 1 to be estimated is only the confirmed category A_N, the value is used as it is for the label.

  FIG. 10B is a diagram illustrating an example of the labeling table 2330. FIG. 10B is an example in which the data of the estimation model construction shaped data 2200 shown in FIG. 7 is labeled using the label shown in FIG. 10A. A data division label 2334 represents a column of labels assigned based on the information of the categorical variable.

  FIG. 10C is a diagram illustrating an example of the labeling table 2340. FIG. 10C is a labeling table 2340 showing the result of assigning data division labels for model construction for estimating the objective variable y1_N + 1 with respect to the tree of FIG. 6B.

  Related categories 2341 and 2342 are variables used for labeling. In this example, there are two types of variables that are the ancestors of y1_N + 1 to be estimated, A_N of the definite category and A_N + 1 of the undetermined category, so an example of labeling with a value using a combination of these two values is shown. ing.

  FIG. 10D is a diagram illustrating an example of the labeling table 2350. The labeling table 2350 includes No 2351, a confirmed category 2352, an unconfirmed category 2353, an estimation target 2354, and a data division label 2355 in one entry.

  FIG. 10D is an example in which the data shown in FIG. 9B is labeled using the data division label 2343 shown in FIG. 10C. A data division label 2355 represents a column of labels assigned based on the information of the categorical variable. These labeling processes are executed in order from the variable corresponding to the node that is far from the root target variable in the tree structure.

  In the example of FIG. 6B, first, the estimation model construction unit 108 executes a labeling process for A_N + 1. In the tree of FIG. 6B, since there is only one A_N ancestor node of A_N + 1, labeling is performed using the value of A_N.

  Next, the estimation model construction unit 108 executes a labeling process for y1_N + 1. In the tree of FIG. 6B, since there are two ancestor nodes of y1_N + 1, A_N + 1 and A_N, the estimation model construction unit 108 labels the value by taking the direct product of the value of A_N + 1 and the value of A_N. The estimation model construction unit 108 stores the labeled information in the data division information storage unit 125.

  In step 305, the estimation model construction unit 108 determines whether or not the estimation model includes an unconstructed variable with respect to the uncertain related category variable read in step 303 of the related category variable reading process. When the uncertain related category variable has not been read or when the estimation model has been constructed for all uncertain related categories, the estimation model construction unit 108 performs the category-specific objective variable estimation model construction processing. Proceed to step 307. On the other hand, if there is an undetermined related category for which the estimation model has not been constructed, the estimation model construction unit 108 proceeds to step 306 of the undetermined category estimation model construction process.

  Step 306 of the uncertain category estimation model construction process is a process executed by the uncertain category estimation model construction unit 110 of the estimation model construction unit 108. The unconfirmed category estimation model construction unit 110 uses the related category variable and tree information read in step 303 of the related category variable reading process and the label information processed in step 304 of the data division labeling process, to determine the unconfirmed category. Build an estimation model for estimating variables.

  At this time, the unconfirmed category estimation model construction unit 110 divides data based on the information labeled in the data division labeling process with respect to the estimation target variable, and constructs an estimation model. For example, when the estimation model for the uncertain category A_N + 1 is constructed using the labeling table shown in FIG. 10A, the uncertain category estimation model construction unit 110 divides the shaped data 201 into data having the same label value. The relationship between the objective variable and the explanatory variable is learned for each data.

  In the first embodiment, as described above, since there are two types of labels, y and n, the uncertain category estimation model construction unit 110 divides the shaped data 201 into two, and estimates models from the respective data. Build up. The estimation model can be constructed using, for example, a known or well-known method such as multinomial logistic regression or BoostingTree.

  In step 307 of the objective variable estimation model construction process, the objective variable estimation model construction section 109 of the estimation model construction section 108 and related category variable and tree information read in step 303 of the related category variable readout process, and data division labeling Using the label information processed in step 304 of the processing, an estimation model for estimating the objective variable is constructed for each related categorical variable value.

  For example, when an estimation model related to the objective variable A_N + 1 is constructed using the labeling table 2300 shown in FIG. 8A, the objective variable estimation model construction unit 109 converts the formatted data 201 into data having the same target label value. And the relationship between the objective variable and the explanatory variable is learned for each data.

  In the first embodiment, since there are two types of labels, y and n, the estimation model construction unit 108 divides the shaped data 201 into two, and constructs an estimation model from each data. For example, when the estimation model construction unit 108 constructs an estimation model related to the objective variable y1_N + 1 (2355) using the labeling table 2330 shown in FIG. 10D, (y, y), (y, n), (n, y) Based on the four types of labels (n, n), the shaped data 201 is divided into four, and an estimation model is constructed from each data. For the construction of the estimation model, widely known methods such as multiple regression, multinomial logistic regression, and BoostingTree can be used.

  The uncertain category estimation model and the objective variable estimation model constructed by the above processing are stored in the estimation model storage unit 126.

  Next, the integrated estimation model construction process performed in step 308 of FIG. 3 will be described below. In the integrated estimation model construction process, first, the future estimation unit 111 performs the process, and then the estimation information aggregation unit 112 performs the process.

  First, the process of the future estimation unit 111 will be described. The future estimation unit 111 performs future estimation of information for each individual (shaped data 201) based on the estimation model stored in the estimation model storage unit 126.

  FIG. 11 is a diagram illustrating a flowchart of processing performed by the future estimation unit 111. Hereinafter, each step will be described.

  In step 1101 of the estimation target information input process, the future estimation unit 111 inputs necessary information such as information on data to be estimated, information on objective variables, information on explanatory variables, and information on models used for estimation. Accept from the unit 102. In addition to the information input from the input unit 102, data stored in advance in the shaped data storage unit 122 may be used as the estimation target data.

  In step 1102 of the model reading process, the future estimation unit 111 reads from the estimation model storage unit 126 the target variable estimation model and the uncertain category estimation model corresponding to the target variable and the explanatory variable to be used.

  In step 1103 of the data division process, the label information for data division stored in the data division information storage unit 125 is read.

  In step 1104, the future estimation unit 111 determines whether there is an unestimated unconfirmed category. This is because whether or not there is an estimation model (hereinafter referred to as an undetermined categorical variable estimation model) for estimating an undetermined categorical variable in the estimation model read by the future estimation unit 111 in step 1101 of the model reading process. Can be determined.

  If there is an undetermined categorical variable and the value of the variable is undetermined, the future estimating unit 111 proceeds to step 1105 of the undetermined category estimating process. On the other hand, if there is no unestimated uncertain categorical variable, the future estimation unit 111 proceeds to step 1106 of the objective variable estimation process.

  In step 1105 of the uncertain category estimation process, the future estimation unit 111 selects and applies one of the uncertain category variable estimation models that can be applied. The application will be implemented in the following two cases.

  The first example will be described. If there is only one uncertain category estimation model, the future estimation unit 111 estimates an uncertain category variable using the uncertain category estimation model. In this case, the unconfirmed category variable satisfies one of the conditions that there is no other related category or that all related categories are confirmed categories.

  If the uncertain category variable does not have a related category, the future estimation unit 111 estimates information on the uncertain category from the explanatory variables using all the shaping data 201. When the uncertain category variable has a related category, the future estimation unit 111 divides the shaped data 201 for each value of the related category, and the uncertain category estimation model corresponding to the value of each category is divided into the divided data. To estimate uncertain categorical variables.

  The second example will be described. When there are two or more uncertain category estimation models, the uncertain category estimation model may include an uncertain category related to the uncertain category.

  For example, there are an uncertain category estimation model X and an uncertain category estimation model Y, the estimation target of the estimation model X is an uncertain category variable C, and the estimation target of the estimation model Y is an uncertain category variable D. Consider a case where the related category included in model Y is C.

  In this case, since the related category C is undetermined, the estimation using the estimation model Y cannot be performed unless the value of C is fixed. Therefore, the future estimation unit 111 first applies the estimation model X to estimate information about the undefined category variable C, and then applies the estimation model Y based on the information of the estimated variable C.

  As described above, the future estimation unit 111 performs the estimation hierarchically by estimating in order from the related category that does not include the uncertain category. Specifically, for example, when the categorical variable C is information that can take two types of values y and n, the future estimation unit 111 first uses the estimation model X to set the probability of y to 0.6. , And the probability value of n is estimated as 0.4.

  Next, the future estimation unit 111 uses the estimation model Y to determine the value Dy of the undefined category variable D when the undefined category variable C is y, and the undefined category variable when the undefined category variable C is n. Two pieces of information of the value Dn of D are estimated.

  Finally, the future estimation unit 111 calculates an expected value from these pieces of information, and sets 0.6 × Dy + 0.4 × Dn as an estimated value of an undetermined categorical variable. Even in the case of estimation using the probability value of the value of the undefined category variable, the future estimation unit 111 uses the data stored in the data division information storage unit 125 to calculate the future estimation unit 111 for data Apply by dividing.

  Next, in step 1106 of the objective variable estimation process, the future estimation unit 111 applies the objective variable estimation model. The objective variable estimation model is applied in the following three cases.

  The first example will be described. When there is no related category in the objective variable estimation model, the future estimation unit 111 estimates information on the objective value from the explanatory variables using all the shaping data 201.

  The second example will be described. When all the related categories included in the objective variable estimation model are definite categorical variables, the future estimation unit 111 divides the shaped data 201 for each category value, and sets the divided data to the uncorresponding value for each category. Estimate uncertain categorical variables by applying a definite category estimation model.

  A third example will be described. When the uncertain categorical variable is included in the related category included in the objective variable estimation model, the estimation is performed using the probability information estimated by the uncertain category variable estimation model. For example, the uncertain category estimation model Y and the objective variable estimation model Z exist, the estimation target of the uncertain category estimation model Y is the uncertain category variable D, the estimation target of the objective variable estimation model Z is the variable z, the purpose Consider a case where the related category included in the variable estimation model Z is the uncertain categorical variable D.

  In this case, since the undetermined categorical variable D is undetermined, estimation using the objective variable estimation model Z cannot be performed unless this value is confirmed. Therefore, the future estimation unit 111 estimates the objective variable using the probability information related to the undefined category variable D estimated using the undefined category estimation model Y. Specifically, for example, the uncertain category variable D is information that can take two types of values y and n, and the probability of the uncertain category variable D estimated by the uncertain category estimation model Y is the probability of y Is a probability value such as 0.3, and the probability of n is 0.7.

  At this time, first, the future estimation unit 111 uses the objective variable estimation model Z to determine the value zy of the objective variable z when the undetermined categorical variable D is y and the objective variable z when the undetermined categorical variable D is n. Two pieces of information of value zn are estimated. Finally, an expected value is calculated from these pieces of information, and 0.3 × zy + 0.7 × zn is set as an estimated value of the objective variable z.

  The estimation information for each individual estimated in this estimation process is stored in the estimation information storage unit 128 together with the shaping data 201.

  Next, processing of the estimation information aggregation unit 112 will be described. The future estimation unit 111 aggregates the individual estimation results stored in the estimation information storage unit 128 based on a specific categorical variable, and generates an estimation result for each group.

  FIG. 14 is a flowchart illustrating an example of processing performed by the estimation information aggregation unit 112. Hereinafter, each step will be described.

  In step 1401 for reading out estimated information, the estimated information aggregating unit 112 reads out information on the estimation result stored in the estimated information storage unit 128.

  In step 1402 for determining the aggregated index, the estimated information aggregating unit 112 determines an index for aggregating the estimation result information read from the estimated information storage unit 128. The estimation information aggregating unit 112 determines, as an index, at least one or more categorical variables included in the shaping data 201 for each individual stored together with the estimation result information as an index to be aggregated. Examples of the index include information obtained by classifying the gender and age included in the basic information for each individual every 10 years, the class of care insurance, and information on the location acquired from the address information.

  In step 1403 for determining the aggregation method, the estimation information aggregating unit 112 determines how to estimate the estimation information for the categorical variable determined in step 1402 for determining the aggregation index. For example, a totaling method such as the sum, average, maximum value, minimum value, median value, mode value, and standard deviation of information aggregated for each categorical variable is determined.

  In step 1404 of the aggregation process, the estimation information aggregating unit 112 aggregates the estimated information by the aggregation method determined in step 1403 of the aggregation method determination using the aggregation index determined in step 1402 of the aggregation index determination. The totaled results are stored in the estimation information storage unit 128.

  An example of a graphical user interface displayed on the output unit 103 by the data analysis unit 101 is shown below.

  FIG. 12 is a diagram of a graphical user interface showing an example of an operation screen for realizing the first embodiment. In the illustrated example, an example of an operation screen 1200 for selecting the shaping data 201, constructing an estimation model of the selected shaping data 201, and displaying information about parameters and architecture of the constructed estimation model 201 is shown.

  In the figure, reference numeral 1201 denotes a button for selecting the shaping data 201. Reference numeral 1202 denotes a button for executing the estimation model construction process. Reference numeral 1203 denotes a button for displaying the parameters and architecture of the estimation model on the screen.

  In the figure, reference numeral 1204 denotes a table that displays information on objective variables, explanatory variables, and categorical variables used in the estimation model.

  In the figure, reference numeral 1205 is a table that displays information on the target variable estimation parameter in the target variable estimation model included in the estimation model for each classification based on the value of the categorical variable. 1206 is a table that displays information on parameters for estimating the uncertain categorical variable in the uncertain categorical variable estimation model included in the estimation model for each classification based on the value of the categorical variable. Reference numeral 1207 denotes a display area for visualizing the relationship among used explanatory variables, objective variables, and categorical variables as a graph of the estimation model architecture.

  A display area 1207 shows an example in which categorical variables are indicated by squares, explanatory variables are indicated by circles, and the relationship between the explanatory variables and the objective variables is indicated by arrows.

  The user of the data analysis unit 101 can select the shaping data 201 necessary for constructing the estimation model by using the operation screen 1200 shown in the first embodiment. Then, the data analysis unit 101 can execute an estimation model construction process based on the selected shaping data. By using the operation screen 1200, the user can grasp the model architecture that is related to various parameters such as explanatory variables of the constructed estimation model, objective variables, and categorical variables.

  FIG. 13 is a diagram of a graphical user interface showing an example of an estimation result display screen 1300 for realizing the first embodiment. The estimation result display screen 1300 shows an example of a screen for displaying an estimation result using an estimation model.

  In the figure, reference numeral 1301 denotes a button for selecting an estimation result. Reference numeral 1302 denotes a table for collecting and displaying the estimated results for each age (or year). In the example of illustration, the example which displays the result of having estimated the medical expenses of type 2 diabetes of 2016 and 2017 based on the information of 2015 is shown.

  In the figure, reference numeral 1303 denotes a pull-down menu for designating the aggregation target, and the estimation information aggregation processing is executed according to the aggregation target designated by the pull-down 1303, and the information in the screen is updated. Reference numeral 1304 denotes a pull-down menu for designating an objective variable of the estimation result to be read. Information on the target disease designated by the pull-down 134 is displayed as information on the screen.

  In the figure, reference numeral 1305 denotes a graph display area for displaying the result of totaling the estimation result of the disease specified by the pull-down 1304, which is the target of aggregation specified by the pull-down 1303. Reference numeral 1306 denotes a map display area for displaying a count result of diseases specified in 1304 as a map, which is specified in 1303.

  Hereinafter, effects of the estimation model construction system according to the first embodiment will be described.

  When the estimation model construction system shown in the first embodiment constructs a model for estimating an objective variable, among the variables included in the explanatory variables, for each categorical variable, the data for which the value of the categorical variable is the same. It is possible to divide the data into two and build an estimation model from each data. Thereby, when estimating the group comprised from the individual from which backgrounds, such as an attribute and past history, differ, estimation accuracy can be improved.

  In the estimation model construction system shown in the first embodiment, when a model is constructed by dividing data based on the value of a categorical variable, whether the value of the categorical variable is a fixed value at the time of estimation or is not yet determined. It is determined whether it is a value.

  The estimation model construction system constructs two models for estimating the categorical variable and two models for estimating the objective variable when the value of the categorical variable is uncertain at the time of estimation. Thereby, even when the value of the categorical variable is uncertain at the time of estimation, it is possible to estimate with high estimation accuracy. For example, when estimating medical costs related to a certain disease in a certain year, there are two models: a model that estimates information on the presence or absence of a disease using information on the presence or absence of a disease in a certain year, and a model that estimates medical costs. By constructing an estimation model and applying them hierarchically, medical costs can be estimated with high accuracy. Moreover, it is possible to estimate both medical expenses and information on the presence or absence of illness at the same time.

  Since the estimation model construction system shown in the first embodiment constructs a model based on the value of the categorical variable, it can grasp the parameters of the model constructed for each value of the categorical variable. Thereby, regarding a specific group, it is possible to grasp from the respective models variables affecting the objective variable of interest.

  The estimation model construction system shown in the first embodiment can aggregate the information aggregated for each individual with an arbitrary categorical variable, and can estimate a group. Thereby, the future health state and medical state of the group can be grasped with a predetermined granularity desired by the user.

  Hereinafter, modifications of the estimation model construction system according to the first embodiment will be described.

  In the first embodiment, as an example of the target data, an example using data including receipt information for each individual and specific medical checkup information in time series will be described. Although the example which builds the model estimated in a year unit was demonstrated, the data which can be handled by the present Example 1 are not limited to this.

  For example, it is possible to widely handle information related to health, medical care or nursing care such as sensing information related to health information and information related to insurance. In addition, a wide variety of time interval models such as quarterly, monthly, etc. can be constructed as time intervals for collecting information. Further, the estimation model construction system according to the first embodiment can be applied to the education field based on the results of periodic tests.

  In the first embodiment, various estimation models can be handled as the estimation model. For example, as models for estimating categorical variable information, models that describe the relationship between explanatory variables and objective variables such as logistic regression and neural networks, support vector machines, and KNN (k-nearest neighbor algorithm) A model that maintains a subset of model building data as a parameter, a model that maintains a stochastic dependency between explanatory variables and objective variables, such as a Bayesian network, and a random forest Various models such as a model in which classification rules are aggregated can be used as an estimation model.

  In the first embodiment, the method of using the related category matrix 501 has been described as the method for specifying the categorical variable related to the objective variable. However, the categorical variable related to the objective variable is read using a method other than the related category matrix 501. Also good. For example, a tree structure showing the relationship between the target variable and related category variable group constructed in the related category variable reading process is built and stored in advance, and the relevant category variable is read out directly using the information. Also good.

  In the first embodiment, an estimation model for estimating for each individual is constructed, and the estimation result is estimated as a group based on a predetermined category. However, the estimation may be performed with a granularity larger than that of the individual. . For example, an estimation model for estimating information aggregated in units of cities may be constructed.

  As described above, the estimation model construction system according to the first embodiment constructs an estimation model construction process for each categorical variable and an estimation model construction process for estimating an undetermined categorical variable at the time of estimation. It is possible to construct an estimation model that enables future estimation with high accuracy even in a group with a mixture of.

  In the first embodiment, the example of the estimation model construction system in which the categorical variable related to the target variable to be estimated is stored in the related category storage unit 124 in advance has been described. In the second embodiment, an example of an estimation model construction system that executes processing for determining a categorical variable associated with an objective variable based on the shaping data 201 will be described. Since the configuration, processing, and the like are the same as those in the first embodiment except for the related category associating unit 113, a duplicate description is omitted.

  The related category associating unit 113 uses the shaping data 201 stored in the shaping data storage unit 122 and the information on the objective variable to determine the category variable item related to the objective variable.

  Hereinafter, a method for determining the item of the categorical variable will be described.

  The first example will be described. Information that is the same item as the objective variable and acquired in the past time series is set as a categorical variable related to the objective variable. For example, in the case where the objective variable is whether or not there is a illness visit in a certain year, the presence or absence of the same illness in the past year is designated as a categorical variable related to the objective variable.

  For example, there is a relationship that whether or not there is a future medical checkup depends greatly on the current medical checkup status. Therefore, the estimation accuracy is improved by separating the estimation model in accordance with the current medical condition of the disease.

  A second example will be described. A variable having the same timing as the objective variable and having a high similarity to the objective variable is selected as a categorical variable. For example, an item whose correlation or mutual information amount is constant (threshold) or more is selected as a categorical variable. For example, as a categorical variable related to medical expenses for a certain disease in the same year, information on the presence or absence of the same disease is designated as a categorical variable. Thereby, the estimation accuracy of the variable depending on other variables can be improved.

  A third example will be described. A certain variable is set as a temporary related categorical variable, and with respect to the variable, an estimation model for estimating an objective variable is constructed from the shaped data 201 divided for each piece of data having the same value of the related categorical variable. A result of estimating the target variable using the result, a result of estimating with the single estimation model without dividing the shaping data 201, or a result of estimating with the estimation model constructed by dividing the shaping data 201 with different categorical variables, One of them is compared, and the related categorical variable is determined based on the comparison result (estimation accuracy) of the estimation. The comparison of the estimation accuracy can be performed by, for example, cross validation using the past shaping data 201.

  Hereinafter, effects of the estimation model construction system according to the second embodiment will be described.

  The estimation model construction system shown in the second embodiment can determine a division method of the shaping data 201 that is a target of estimation model construction based on the shaping data 201. Thereby, for example, even when the quality of information included in the shaped data 201 is changed, it is possible to automatically construct an appropriate estimation model for the shaped data 201.

  In addition, this invention is not limited to an above-described Example, Various modifications are included. For example, the above-described embodiments are described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described. Further, a part of the configuration of one embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of one embodiment. In addition, any of the additions, deletions, or substitutions of other configurations can be applied to a part of the configuration of each embodiment, either alone or in combination.

  Each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. In addition, each of the above-described configurations, functions, and the like may be realized by software by the processor interpreting and executing a program that realizes each function. Information such as programs, tables, and files for realizing each function can be stored in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.

  Further, the control lines and information lines indicate what is considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. Actually, it may be considered that almost all the components are connected to each other.

DESCRIPTION OF SYMBOLS 101 Data analysis part 102 Input part 103 Output part 104 Arithmetic apparatus 105 Memory 106 Storage medium 107 Data shaping part 108 Estimate model construction part 109 Objective variable estimation model construction part 110 Uncertain category estimation model construction part 111 Future estimation part 112 Estimation information aggregation Unit 113 related category association unit 120 database 121 data storage unit 122 shaped data storage unit 124 related category storage unit 125 data division information storage unit 126 estimation model storage unit 127 estimation model construction variable group storage unit 128 estimation information storage unit

Claims (15)

An estimation model construction system comprising a processor for executing a program and a memory for storing the program, and constructing an estimation model by executing the program,
Stored information on medical and health for each individual, related category information for storing correspondence between data items included in the stored information, and other data items related to the data item and data items including category information And a storage unit for storing
Based on the received information or the accumulated information, the first objective variable that is the estimation target of the estimation model and the explanatory variable that is the input information of the estimation model are determined, and based on the correspondence relationship of the related category information, A categorical variable corresponding to the first objective variable is determined, the accumulated information whose value satisfies the predetermined condition is divided, and a first estimation model corresponding to each divided accumulated information is constructed. An estimation model construction unit to
The future in which the first objective variable, the explanatory variable, and the categorical variable are information to be estimated, and the first estimation model to be applied is selected according to the value of the categorical variable included in the information to be estimated An estimation section;
The estimation model construction system characterized by having. An estimation model construction system according to claim 1,
The estimation model construction unit further includes:
It is determined whether or not the information on the categorical variable is included in the information to be estimated. If the information is not included in the information on the estimation target, the categorical variable is set as a second objective variable, and the explanatory variable A second estimation model for estimating the probability distribution of the value of the categorical variable based on
The future estimation part is
Information on the probability distribution of the value of the categorical variable, which is the second objective variable, is estimated based on the explanatory variable using the second estimation model, and further, the probability distribution of the value of the categorical variable thus estimated is estimated. An estimation model construction system for estimating the first objective variable by applying the first estimation model constructed for each stored information in which the value of the categorical variable satisfies a predetermined condition according to information . An estimation model construction system according to claim 1,
The estimation model construction unit
By repeating the process of extracting a categorical variable related to the first objective variable that is the estimation target of the estimation model and the process of using the extracted categorical variable as a new first objective variable, the first Extract a tree structure having one objective variable as a root node and an internal node of a categorical variable related to the first objective variable;
Based on the tree structure, build an estimation model with the root node or internal node as a first objective variable,
Further, when building an estimation model with each node as a first objective variable, the accumulated information is divided for each piece of information in which the value of a categorical variable that is a child node of the internal node satisfies a predetermined condition, Build an estimation model with each node as the first objective variable for each divided stored information,
The future estimation part is
Estimating an estimation model of a categorical variable corresponding to the internal node using an estimation model having the internal node as a first objective variable, and further, if the internal node has child nodes, By applying the categorical variable estimation model corresponding to the internal node according to the information on the probability distribution of the value of the categorical variable corresponding to the child node estimated from the categorical variable estimation model as the first objective variable, the internal variable An estimation model construction system characterized by estimating information on probability distribution of nodes. An estimation model construction system according to claim 1,
An estimation model construction system, further comprising a related category associating unit that determines an item of a categorical variable related to the first objective variable using the accumulated information and information of the first objective variable. An estimation model construction system according to claim 2,
The estimation model has a parameter for estimating the first objective variable,
The estimation model construction system further includes at least one of the parameter of the estimation model, the relationship between the first objective variable, the explanatory variable, and the categorical variable related to the first objective variable included in the estimation model. An estimation model construction system comprising an output unit for presentation. An estimation model construction method for constructing an estimation model with a computer having a processor and a memory,
The computer stores correspondence information between stored information on medical care and health for each individual, data items included in the stored information, and other data items related to the data item and data items including category information A first step of storing the related category information in the storage unit of the computer,
A second step in which the computer determines a first objective variable that is an estimation target of the estimation model and an explanatory variable that is input information of the estimation model based on the received information or the accumulated information;
A third step in which the computer determines a categorical variable corresponding to the first objective variable based on a correspondence relationship of the related category information;
A fourth step in which the computer divides the stored information in which the value of the categorical variable satisfies a predetermined condition, and constructs a first estimation model corresponding to each of the divided stored information;
The computer uses the first objective variable, the explanatory variable, and the categorical variable as information to be estimated, and applies the first estimation model according to the value of the categorical variable included in the information to be estimated. A fifth step of selecting
The estimation model construction method characterized by including. An estimation model construction method according to claim 6,
The fourth step further includes:
It is determined whether or not the information on the categorical variable is included in the information to be estimated. If the information is not included in the information on the estimation target, the categorical variable is set as a second objective variable, and the explanatory variable A second estimation model for estimating the probability distribution of the value of the categorical variable based on
The fifth step includes
Information on the probability distribution of the value of the categorical variable, which is the second objective variable, is estimated based on the explanatory variable using the second estimation model, and further, the probability distribution of the value of the categorical variable thus estimated is estimated. An estimation model construction method for estimating the first objective variable by applying the first estimation model constructed for each accumulated information in which the value of the categorical variable satisfies a predetermined condition according to information . An estimation model construction method according to claim 6,
The fourth step includes
By repeating the process of extracting a categorical variable related to the first objective variable that is the estimation target of the estimation model and the process of using the extracted categorical variable as a new first objective variable, the first A tree structure having one objective variable as a root node and an internal node of a categorical variable related to the first objective variable is extracted, and based on the tree structure, the root node or the internal node is defined as a first objective variable. When each estimation model is constructed, and further, when the estimation model having each node as the first objective variable is constructed, the value of the categorical variable that is a child node of the internal node satisfies each predetermined condition. Dividing the accumulated information, constructing an estimation model with each node as a first objective variable for each of the divided accumulated information,
The fifth step includes
Estimating an estimation model of a categorical variable corresponding to the internal node using an estimation model having the internal node as a first objective variable, and further, if the internal node has child nodes, By applying the categorical variable estimation model corresponding to the internal node according to the information on the probability distribution of the value of the categorical variable corresponding to the child node estimated from the categorical variable estimation model as the first objective variable, the internal variable An estimation model construction method characterized by estimating node probability distribution information. An estimation model construction method according to claim 6,
A method for constructing an estimation model, further comprising a sixth step of determining an item of a categorical variable related to the first objective variable using the accumulated information and information of the first objective variable. An estimation model construction method according to claim 7,
The estimation model has a parameter for estimating the first objective variable,
A seventh step of presenting at least one of the parameters of the estimation model, the relationship between the first objective variable, the explanatory variable, and the categorical variable related to the first objective variable included in the estimation model; The estimation model construction method characterized by including further. A program for controlling a computer having a processor and a memory,
Stored information on medical and health for each individual, related category information for storing correspondence between data items included in the stored information, and other data items related to the data item and data items including category information And a first step of storing in the storage unit of the computer,
A second step of determining a first objective variable that is an estimation target of the estimation model based on the received information or the accumulated information, and an explanatory variable that is input information of the estimation model;
A third step of determining a categorical variable corresponding to the first objective variable based on a correspondence relationship of the related category information;
A fourth step of dividing the accumulated information in which the value of the categorical variable satisfies a predetermined condition, and constructing a first estimation model corresponding to each of the divided accumulated information;
The first objective variable, the explanatory variable, and the categorical variable are used as estimation target information, and the first estimation model to be applied is selected in accordance with the value of the categorical variable included in the estimation target information. 5 steps,
Is executed by the computer. The program according to claim 11,
The fourth step further includes:
It is determined whether or not the information on the categorical variable is included in the information to be estimated. If the information is not included in the information on the estimation target, the categorical variable is set as a second objective variable, and the explanatory variable A second estimation model for estimating the probability distribution of the value of the categorical variable based on
The fifth step includes
Information on the probability distribution of the value of the categorical variable, which is the second objective variable, is estimated based on the explanatory variable using the second estimation model, and further, the probability distribution of the value of the categorical variable thus estimated is estimated. A program for estimating the first objective variable by applying the first estimation model constructed for each stored information in which the value of the categorical variable satisfies a predetermined condition according to information. The program according to claim 11,
The fourth step includes
By repeating the process of extracting a categorical variable related to the first objective variable that is the estimation target of the estimation model and the process of using the extracted categorical variable as a new first objective variable, the first A tree structure having one objective variable as a root node and an internal node of a categorical variable related to the first objective variable is extracted, and based on the tree structure, the root node or the internal node is defined as a first objective variable. When each estimation model is constructed, and further, when the estimation model having each node as the first objective variable is constructed, the value of the categorical variable that is a child node of the internal node satisfies each predetermined condition. Dividing the accumulated information, constructing an estimation model with each node as a first objective variable for each of the divided accumulated information,
The fifth step includes
Estimating an estimation model of a categorical variable corresponding to the internal node using an estimation model having the internal node as a first objective variable, and further, if the internal node has child nodes, By applying the categorical variable estimation model corresponding to the internal node according to the information on the probability distribution of the value of the categorical variable corresponding to the child node estimated from the categorical variable estimation model as the first objective variable, the internal variable A program characterized by estimating the probability distribution information of a node. The program according to claim 11,
A program, further comprising a sixth step of determining an item of a categorical variable related to the first objective variable using the accumulated information and information of the first objective variable. A program according to claim 12,
The estimation model has a parameter for estimating the first objective variable,
A seventh step of presenting at least one of the parameters of the estimation model, the relationship between the first objective variable, the explanatory variable, and the categorical variable related to the first objective variable included in the estimation model; Furthermore, the program characterized by including.

Images