JP6937359B2 - Cluster division evaluation device, cluster division evaluation method and cluster division evaluation program - Google Patents

Description

The present invention relates to a cluster division evaluation device, a cluster division evaluation method, and a cluster division evaluation program.

For example, when a company plans to open a new store, the company predicts where, what size of store, and how much sales can be expected. In addition, other companies predict the purchase amount (total purchasing power) of a customer based on the data of the customer who purchased a certain product. In this way, there are many opportunities to predict the purchase amount of the buyer or the sales of the seller for marketing. As a method for making such a prediction, a regression analysis using past data of a store or a customer is known. The characteristic value predictor of Patent Document 1 also uses a regression analysis method to predict characteristic values (objective variables) in a wide range of fields including sales management.

The characteristic value predictor of Patent Document 1 creates a combination of one variable, a combination of two variables, a combination of three variables, ..., A combination of n variables from n candidates for explanatory variables. do. The total number of these combinations is _n C ₁ + _n C ₂ + _n C ₃ + ... + _n C _n . The prediction device creates a prediction model using the measured values of the explanatory variables belonging to each combination and the measured values of the objective variable at that time. Then, the prediction device obtains the difference (error) between the predicted value of the objective variable output by the prediction model and the measured value, and sets the logarithm of the variance of the error as a "reference value". The prediction device calculates reference values in order from the combination with the smallest number of explanatory variables. Then, the reference value rises as the number of explanatory variables increases. The prediction device sets the combination immediately before the degree of increase becomes smaller than a predetermined threshold value as the combination of explanatory variables for predicting the objective variable.

Japanese Unexamined Patent Publication No. 7-93284

In forecasting sales or purchase amount, data may be divided into multiple categories according to the characteristics of the data. For example, the categories in the sales forecast of large-scale retail stores are "city center", "residential area", "suburbs", and the like. However, there are categories that are often overlooked from ordinary business knowledge. For example, the northwestern part of the Imperial Palace in Tokyo is an "office district" and a "luxury residential district," although it is a "city center." The beach area of Chiba Prefecture is an "office district" and a "high-rise condominium district," although it is a "suburb." In either case, the general categories "city center", "residential area" and "suburbs" cannot be used.

The characteristic value prediction device of Patent Document 1 is for creating an optimum prediction model, but it focuses attention on the selection of explanatory variables and lacks the idea of using the prediction model properly according to the characteristics of the data. ..
Therefore, an object of the present invention is to divide a multidimensional space of variables so that a plurality of highly accurate prediction models can be used according to the characteristics of data.

The cluster division evaluation device of the present invention includes a survey value acquisition unit that acquires the survey value of the attribute of the trading entity and the survey value of the economic power of the trading entity, and the survey value of a plurality of attributes in the trading entity and the economy of the trading entity. Based on the point showing the power survey value, the clustering part that divides into multiple clusters in the multidimensional space, and the attribute of the trading entity is explained based on the attribute survey value and economic power survey value of each divided cluster. Create a prediction model with the economic power of the trading entity as a variable , optimize the parameters of the prediction model for each of multiple clusters, and predict the economic power of the trading entity output by the prediction model with optimized parameters. It is characterized by including a regression analysis unit that evaluates the difference between the value and the survey value of the economic power of the trading entity for each number of clusters and determines the number of clusters based on the evaluation result.
Other means will be described in the form for carrying out the invention.

According to the present invention, the multidimensional space of variables can be divided so that a plurality of highly accurate prediction models can be used according to the characteristics of the data.

It is a figure explaining the structure of the cluster division evaluation apparatus. This is an example of survey value information. This is an example of cluster information. It is a figure explaining the relationship between a cluster and a prediction model. It is a figure explaining the relationship between a cluster and a prediction model. It is a figure explaining the relationship between a cluster and a prediction model. It is a figure explaining the relationship between a cluster and a prediction model. It is a figure explaining an error. This is an example of error information. It is a flowchart of a processing procedure. This is an example of survey value information.

Hereinafter, a mode for carrying out the present invention (referred to as “the present embodiment”) will be described in detail with reference to figures and the like. This embodiment has a first embodiment and a second embodiment. The first embodiment is an example of predicting the annual sales of a store, and the second embodiment is an example of predicting the annual purchase amount of a customer. These are just examples, and the present invention is more generally applicable to marketing quantity forecasting.

<First Embodiment>
First, the first embodiment will be described.

(Cluster division evaluation device)
The configuration of the cluster division evaluation device 1 and the like will be described with reference to FIG. The cluster division evaluation device 1 is a general computer, and includes a central control device 11, an input device 12 such as a mouse and a keyboard, an output device 13 such as a display, a main storage device 14, an auxiliary storage device 15, and a communication device 16. .. These are connected to each other by a bus. The auxiliary storage device 15 stores the prediction model 31, the survey value information 32, the cluster information 33, and the error information 34 (details will be described later).

The survey value acquisition unit 21, the clustering unit 22, the regression analysis unit 23, and the display processing unit 24 in the main storage device 14 are programs. The central control device 11 reads these programs from the auxiliary storage device 15 and loads them into the main storage device 14, thereby realizing the functions of the respective programs (details will be described later). The auxiliary storage device 15 may have a configuration independent of the cluster division evaluation device 1.

The store server 3 and the card company server 4 are connected to the cluster division evaluation device 1 via the network 2. The cluster division evaluation device 1 can acquire various data related to the store and the customer from the store server 3 and the card company server 4.

(Prediction model)
The prediction model 31 of this embodiment is a linear equation such as the following equation 1.
y = a ₀ + a ₁ x ₁ + a ₂ x ₂ + a ₃ x ₃ (Equation 1)

Here, y is the annual sales of the store. x ₁ is the sales floor area of the store. x ₂ is the property tax land price of the land where the store is located. x ₃ is the number of parking lots in the store. a ₀ , a ₁ , a ₂ and a ₃ are constants (parameters). Equation 1 is a function with x ₁ , x ₂ and x ₃ as explanatory variables and y as an objective variable. Then, _{by changing the values of a 0} , a ₁ , a ₂ and a ₃ in various ways, the shape of the prediction model 31 and its position in the four-dimensional space are changed in various ways. Here, the number (type) of variables is set to "4", which is just an example. The number of variables may be larger, that is, the dimensions of the prediction model may be larger.

Now, suppose that there are many _{"[Y, X 1} , X ₂ , X ₃ ]" as a combination of past survey values of annual sales, sales floor area, property tax route price, and number of parking lots. The types of numerical values indicated by each of Y, X ₁ , X ₂ and X ₃ are the same as the types of numerical values indicated by each of _{y, x 1} , x ₂ and x _3. However, for convenience of explanation, the survey values as actual examples are shown in uppercase, and the variables of the prediction model are shown in lowercase. The output (objective variable) y of the prediction model is a “prediction value”. "Y-y" is called an error. The cluster division evaluation device 1 uses a combination of survey values to minimize a combination of parameters “[a ₀ , a ₁ , a ₂ , a ₃ ^{” that minimizes the sum of squares “Σ (Y−y) 2” of the error.} ] ”Is determined (details below).

(Survey value information)
FIG. 2 is an example of the survey value information 32. In the survey value information 32, the survey value of the objective variable is stored in the objective variable column 102 and the survey value of the explanatory variable is stored in the explanatory variable column 103 in association with the store ID stored in the store ID column 101. ..
The store ID in the store ID column 101 is an identifier that uniquely identifies the store that is the transaction entity.

The survey value of the objective variable in the objective variable column 102 is the annual sales of the store. “Annual” is just an example, and the survey value of the objective variable may be monthly sales, weekly sales, or sales in other periods. “#” Indicates a different value abbreviated (the same applies hereinafter).

The survey values of the explanatory variables in the explanatory variable column 103 are the sales floor area (column 103a), the property tax route price (column 103b), and the number of parking lots (column 103c).
Of these, the sales floor area is the area of the total floor area of the store that is directly used for selling products to customers.
The property tax land price is the land price that is the basis for calculating the property tax levied on the land where the store is located.
The number of parking lots is the number of vehicles that can be accommodated in the parking lot that can be used by customers who visit the store.
The annual sales correspond to the "economic power of the transaction entity", and the sales floor area, property tax route price and the number of parking lots correspond to the "attributes of the transaction entity".

(Clustering)
Grouping many points drawn in a multidimensional space by those with similar positions is generally called "clustering". As a clustering method, the "k-means method" is well known. The cluster division evaluation device 1 also uses the k-means method as described in (1) to (5) below.

(1) The cluster division evaluation device 1 appropriately assigns each of the plurality of points to any of the k groups.
(2) Clustering evaluation apparatus 1 calculates the sum of squares d _i of the distance from the gravity center of a group to the point belonging to the group for each group. i is a group number (i = 1, 2, ..., K).

(3) The cluster division evaluation device 1 calculates _{D k} = _Σ di after changing the affiliation of one point from a certain group to another group. D _k is the sum of _{d i} of the k groups. The cluster division evaluation device 1 repeats the process for each combination of the point of changing the affiliation and the new affiliation.
(4) The cluster division evaluation device 1 determines the affiliation of each point so as to minimize _{D k.}
(5) The cluster division evaluation device 1 repeats the processes (1) to (4) after changing k to 1, 2, 3, ....

(Cluster information)
FIG. 3 is an example of cluster information 33. In the cluster information 33, in association with the store ID stored in the store ID column 111, the objective variable column 112 contains the survey value of the objective variable, the explanatory variable column 113 contains the survey value of the explanatory variable, and the cluster ID column to which the objective variable belongs. The belonging cluster ID is stored in 114.

The store ID in the store ID column 111 is the same as the store ID in FIG.
The survey value of the objective variable in the objective variable column 112 is the same as the survey value of the objective variable in FIG.
The survey value of the explanatory variable in the explanatory variable column 113 is the sales floor area (column 103a) among the survey values of the explanatory variable in FIG. For the sake of simplification of the explanation, the explanatory variable in FIG. 3 is only "sales floor area".

The affiliation cluster ID column 114 is divided into a cluster number 1 column 114a, a cluster number 2 column 114b, a cluster number 3 column 114c, a cluster number 4 column 114d, and so on, according to the number of clusters. The cluster ID is stored in each of the separated columns. The cluster ID is an identifier that uniquely identifies the cluster. Each cluster corresponds to the regional characteristics of the store. In general, a company planning to open a new store determines various numerical values (store opening patterns) including annual sales according to regional characteristics. Even if the points to which they belong are the same as in "c3" and "c10", different cluster IDs are assigned if the number of clusters is different ( _Dk is recalculated). Because).

4 to 7 are diagrams for explaining the relationship between the cluster and the prediction model 31. FIG. 4 corresponds to the number of clusters 1 column 114a in FIG. The horizontal axis of the coordinate plane of FIG. 4 is an explanatory variable (sales floor area), and the vertical axis is an objective variable (annual sales). Twenty points ● corresponding to the stores M01 to M20 are drawn on the coordinate plane (the same applies to FIGS. 5 to 7). The circle c1 represents the cluster c1. The straight line 31a represents the prediction model 31 (FIG. 1). The method of creating the prediction model will be described later.

FIG. 5 corresponds to the number of clusters column 2 114b in FIG. The circle c2 represents the cluster c2. The circle c3 represents the cluster c3. The straight line 31b represents the prediction model 31 (FIG. 1). The straight line 31c also represents the prediction model 31 (FIG. 1).
FIG. 6 corresponds to the number of clusters 3 column 114c in FIG. FIG. 7 corresponds to the number of clusters in column 114d in FIG. The description of FIGS. 6 and 7 is based on the description of FIG.
In FIGS. 4 to 7, the center of the circle c1 or the like is not the center of gravity of the cluster c1 or the like (the average of the coordinate values of all the points ●) due to the restrictions on drawing.

In FIGS. 4 to 7, the cluster division evaluation device 1 creates a prediction model for each cluster using only the survey values ● belonging to the cluster. The method by which the cluster division evaluation device 1 _{creates the prediction model “y = a 0} + a ₁ x ₁ ” is as follows (11) to (17).

(11) Clustering evaluation apparatus 1 substitutes random manner the values of the parameters _{a 0} and _{a 1} which is generated in _{a 0} and _{a 1} of _the prediction model.
(12) The cluster division evaluation device 1 _{substitutes the survey value X into x 1} of the prediction model and calculates y.
(13) The cluster division evaluation device 1 calculates an error "Y-y".
(14) The cluster division evaluation device 1 changes the values of [X, Y] for each store and repeats the processes (12) and (13).

(15) The cluster division evaluation device 1 calculates "Σ (Y-y) ^{2 " which is the sum of "(Y-y) 2" of each store.}
(16) The cluster division evaluation device 1 substitutes the other values of _{the parameters a 0} and a _{1 randomly generated into the prediction models a 0} and a _1, and then substitutes the values (12) to (15). Is repeated a sufficiently large number of times.
(17) The cluster division evaluation device 1 determines the values of the parameters a _0S and a _1S ^{that minimize “Σ (Y−y) 2”.} Here, "S" indicates that it is "optimized".

(error)
FIG. 8 is a diagram illustrating an error. The horizontal axis of the coordinate plane of FIG. 8 is the sales floor area, and the vertical axis is the annual sales. The 20 points ● indicate the combination of survey values [X, Y] in FIG. The straight line 31a is a prediction model 31 (FIG. 1), and the formula is “y = a _0S + a _1S x ₁ ”. An error "Y-y" is defined for each of the points ●. As mentioned above, "Σ (Y-y) ² " is minimized, but when focusing on individual points ●, some have almost no error and some have relatively large error. There is.

(Error information)
FIG. 9 is an example of the error information 34. In the error information 34, an error is stored in the error column 122 and an error evaluation value is stored in the error evaluation value column 123 in association with the number of clusters stored in the cluster number column 121.
The number of clusters in the number of clusters column 121 is the number of clusters.
The error in the error column 122 is “√ (Σ (Y−y) ² / n)”. Here, n is the number of points ● in the cluster. “√ (Σ (Y−y) ² / n)” is the square root of the average of the sum of squares of the errors in FIG. The cluster ID is described in parentheses attached to "#".

The error evaluation value in the error evaluation value column 123 is an arbitrary value obtained by processing the error, and the smaller the value, the higher the evaluation for the number of clusters. The error evaluation value is, for example, the average of the errors included in the record (row) of the error information 34, the minimum value of the error, the variance of the error, and the like. Depending on how the error evaluation value is defined, the larger the value, the higher the evaluation for the number of clusters.

(Processing procedure)
FIG. 10 is a flowchart of the processing procedure. As a premise for starting the processing procedure, it is assumed that the survey value information 32 (FIG. 2) is stored in the auxiliary storage device 15 in a completed state.
In step S201, the survey value acquisition unit 21 of the cluster division evaluation device 1 acquires the survey value. Specifically, the survey value acquisition unit 21 acquires the survey value information 32 (FIG. 2) from the auxiliary storage device 15.

In step S202, the clustering unit 22 of the cluster division evaluation device 1 receives the variable. Specifically, the clustering unit 22 accepts a user to select a part or all of a plurality of explanatory variables via the input device 12. For example, the value a ₁ of the parameters of floor space among users explanatory variable is largest will likely of all the parameters except a _0, i.e., floor space is likely to be the largest influence on the objective variable If expected, the user may select "sales floor area". Here, it is assumed that the user selects "sales floor area".

In step S203, the clustering unit 22 receives the maximum value of the number of clusters and the like. Specifically, the clustering unit 22 selects the minimum and maximum values of the number of clusters and the minimum value of the points ● (the number of records of the cluster information 33) included in one cluster via the input device 12. Accept to do. Here, it is assumed that the user inputs "1" as the minimum value of the number of clusters, "4" as the maximum value of the number of clusters, and "4" as the minimum value of the points ● included in one cluster.

In step S204, the clustering unit 22 performs clustering. Specifically, first, the clustering unit 22 deletes the columns of explanatory variables other than the "sales floor area" from the survey value information 32 (FIG. 2).
Secondly, the clustering unit 22 uses the k-means method described above to obtain 20 points of the survey value information 32 (FIG. 2) ● “[X, Y] = [sales floor area, annual sales]”. Is divided into k clusters (k = 1, 2, 3, 4). At this time, the clustering unit 22 ensures that each cluster contains at least four points ●.

In step S205, the clustering unit 22 creates the cluster information 33 (FIG. 3). Specifically, the clustering unit 22 creates the cluster information 33 based on the result of clustering in the “second” step S204.

In step S206, the regression analysis unit 23 of the cluster division evaluation device 1 creates the prediction model 31. Specifically, the regression analysis unit 23 accepts the user to describe the mathematical formula of the prediction model on the screen, or displays the template of the general prediction model on the screen and accepts the user to select. The prediction model 31 created here does not have to be a linear equation as in Equation 1 described above, and may be a higher-order equation or a non-linear equation including an exponent, a logarithm, and the like. .. However, it is assumed that the prediction model 31 includes parameters (values are unknown at this stage) for each variable received in step S202.

In step S207, the regression analysis unit 23 optimizes the parameters for each cluster. Specifically, the regression analysis unit 23 determines the parameters of the prediction model for each cluster by the method described above. That is, the regression analysis unit 23 uses the survey values of the stores M01 to M20 of the survey value information 32 (FIG. 2) that belong to the cluster to be processed, and minimizes ^{"Σ (Y-y) 2".} Determine the parameters to be used.

In step S208, the regression analysis unit 23 creates the error information 34 (FIG. 9). Specifically, first, the regression analysis unit 23 creates the error information 34. The error information 34 created here has four records, and "1", "2", "3", and "4" are stored in the cluster number column 121. The error column 122 and the error evaluation value column 123 are blank.
Second, the regression analysis unit 23 calculates the error “√ (Σ (Y−y) ² / n)” using the minimum “Σ (Y−y) ^{2” in step S207.} It is stored in the error column 122.
Thirdly, the regression analysis unit 23 calculates an error evaluation value based on the error of each record and stores it in the error evaluation value column 123.

In step S209, the regression analysis unit 23 determines the number of clusters based on the error evaluation value. Specifically, the regression analysis unit 23 determines the number of clusters of records having the smallest error evaluation value. Here, “minimum” is merely an example, and the regression analysis unit 23 may determine a plurality of “clusters” whose error evaluation values are small enough to satisfy a predetermined criterion.

In step S210, the display processing unit 24 of the cluster division evaluation device 1 displays the determined number of clusters and the error evaluation value. Specifically, first, the display processing unit 24 displays the number of clusters determined in step S209 and the error evaluation value for the number of clusters on the output device 13. Here, it is assumed that "the number of clusters = 4" is displayed.
Second, the display processing unit 24 stores the prediction models 31g, 31h, 31i and 31j (FIG. 7) corresponding to the four clusters c7, c8, c9 and c10 in the auxiliary storage device 15. After that, the processing procedure ends.

(Utilization of prediction model)
Assuming that "number of clusters = 4" is displayed in "first" of step S210, a method of utilizing the prediction model after that will be described. The cluster c7 corresponds to the stores M01 to M04 in FIG. Stores M01 to M04 are, for example, stores located in a specific area. The regression analysis unit 23 uses the prediction model 31g when predicting the annual sales of stores newly opened in the area. The cluster c10 corresponds to the stores M17 to M20 in FIG. The stores M17 to M20 are, for example, stores located in a specific area. The regression analysis unit 23 uses the prediction model 31j when predicting the annual sales of stores newly opened in the other area. The same applies to other clusters.

(Modified example of processing procedure)
In the above, the clustering unit 22 clusters for all the number of clusters, and the regression analysis unit 23 calculates the error evaluation value for all the number of clusters (round-robin processing). However, the clustering unit 22 may perform clustering in descending or ascending order of the number of clusters k = 1, 2, 3, and 4, and the regression analysis unit 23 may repeat the process of calculating the error evaluation value. In this case, the clustering unit 22 and the regression analysis unit 23 repeat the process until the error evaluation value reaches a predetermined threshold value (target) or the amount of decrease in the error evaluation value from the previous time becomes equal to or less than the predetermined threshold value.

<Second embodiment>
Subsequently, the second embodiment will be described. The difference between the second embodiment and the first embodiment is that instead of the first embodiment using the survey value information 32 (FIG. 2), the second embodiment uses the survey value information 32b (FIG. 2). 11) is to be used.

FIG. 11 is an example of the survey value information 32b. In the survey value information 32b, the survey value of the objective variable is stored in the objective variable column 132 and the survey value of the explanatory variable is stored in the explanatory variable column 133 in association with the customer ID stored in the customer ID column 131. ..
The customer ID in the customer ID column 131 is an identifier that uniquely identifies the customer who is the legal entity.

The survey value of the objective variable in the objective variable column 132 is the annual purchase amount of the customer. The annual purchase amount here is the total price of the products purchased at all stores. “Annual” is just an example, and the survey value of the objective variable may be the monthly purchase amount or the purchase amount for other periods. “#” Indicates a different value abbreviated (the same applies hereinafter).
The survey values of the explanatory variables in the explanatory variable column 133 are age (column 133a), gender (column 133b), self-owned (column 133c), annual income (column 133d), payment method (column 133e), and food purchase amount (column 133f). ) And the purchase price of clothing (column 133 g).

Of these, the age is the age of the customer.
Gender is the gender of the customer. In FIG. 11, for the sake of clarity, the gender is either “male” or “female”. When gender is assigned to the axis of the explanatory variable of the multidimensional space, it is quantified as, for example, "male = 0, female = 1" (the same applies to the self-owned and payment methods described later).

Self-ownership is either "yes", which indicates that the customer owns the home in which the customer resides, or "none", which indicates that the customer does not own the home. If self-owned is "yes", for example, the area or market value of the land may be stored in this column.
Annual income is the amount of annual income of the customer. The customer purchases the product or the like at the amount obtained by subtracting the savings amount or the repayment amount from the annual income.
The payment method is either "cash" indicating that the product price has been paid in cash, or "card" indicating that the product has been paid in cash.

The food purchase amount is the annual total amount of the food price of the products purchased at all stores.
The clothing purchase amount is the annual total amount of clothing purchases among the products purchased at all stores.
“Food” and “clothing” are just examples. The user can select the purchase amount of a specific item that is likely to correlate with the product or service to be analyzed as the use of the annual purchase amount. For example, it is well known that the ratio of food purchase amount to annual purchase amount (Engel coefficient) is almost constant for each social class.
The annual purchase amount corresponds to the "economic power of the transaction entity", and the age, gender, self-owned property, annual income, payment method, food purchase amount and clothing purchase amount correspond to the "attribute of the transaction entity".

The record of the survey value information 32b is basically stored for each customer. However, it is also possible to make one record correspond to one purchase opportunity (payment unit) and allow duplication for each customer. When records are stored on a customer-by-customer basis, a single record typically reflects multiple purchasing opportunities for that customer. Therefore, "card" and "cash" as payment methods may be mixed. In this case, as the payment method, the one with the larger number of payments or the one with the larger amount of money may be typically adopted. Further, a two-dimensional value such as "card = # times, cash = # times" or "card = # yen, cash = # yen" may be adopted.

The specific processing content of the second embodiment is the same as that of the first embodiment, and the description of FIGS. 1 and 3 to 10 applies to the second embodiment as it is. However, in FIG. 3, "store ID" (column 111) is read as "customer ID". “Annual sales” (column 112) should be read as “annual purchase amount”. The “sales floor area” (column 113) should be read as any one of columns 133a to 133g in FIG. 11 (for example, “annual income”).

In the second embodiment as in the first embodiment, on the premise that "the number of clusters = 4" is displayed in the "first" of step S210, a method of utilizing the prediction model thereafter will be described. The cluster c7 corresponds to the customers P01 to P04 in FIG. 3 after the replacement. Customers P01 to P04 are, for example, customers who live in a specific area. The regression analysis unit 23 uses the prediction model 31g when predicting the annual purchase amount of a customer residing in the area. The cluster c10 corresponds to the customers P17 to P20 of FIG. Customers P17 to P20 are, for example, customers who live in a specific area. The regression analysis unit 23 uses the prediction model 31j when predicting the annual purchase amount of a customer residing in the other area. The same applies to other clusters.

Through the first embodiment and the second embodiment, examples of stores and customers who are the main constituents of commodity transactions have been described. However, as is clear from the above, the present invention can also be applied to a legal entity other than the customer of the buyer of the goods and the store of the seller of the goods. Transaction entities include, for example, lenders and borrowers in goods lease transactions, creditors and debtors in capital transactions, service providers and recipients in transportation, logistics, accommodation, medical care, education, nursing care services, and the like.

An example of forecasting annual sales and annual purchase amount has been described through the first embodiment and the second embodiment. However, as is clear from the above, the present invention can be applied to an example of predicting the general economic power of a legal entity. Economic power includes loan balance, borrowing balance, quantity of goods or services provided or provided, market share, etc.

Through the first embodiment and the second embodiment, an example in which the explanatory variables of the store and the customer, which are the subjects of the product transaction, are the data that can be acquired from the store server 3 or the card company server 4 has been described. However, as is clear from the above, the present invention can be applied to any attribute of a legal entity that can be quantified.

(Effect of this embodiment)
The effects of the cluster division evaluation device of this embodiment are as follows.
(1) The cluster division evaluation device can create a highly accurate prediction model for each attribute of the legal entity.
(2) The cluster division evaluation device can display the expected error evaluation value and the number of clusters corresponding to the attributes of the legal entity.
(3) The cluster division evaluation device enables the user to specify the number and size of clusters.
(4) The cluster division evaluation device can be applied to forecast the sales of a store or the purchase amount of a customer.
(5) The cluster division evaluation device can use the attributes of the legal entity that are generally easily available.

The present invention is not limited to the above-described embodiment, and includes various modifications. For example, the above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to the one including all the configurations described. Further, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. Further, it is possible to add / delete / replace a part of the configuration of each embodiment with another configuration.

Further, each of the above-mentioned configurations, functions, processing units, processing means and the like may be realized by hardware by designing a part or all of them by, for example, an integrated circuit. Further, each of the above 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 that realize each function can be placed 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.
In addition, control lines and information lines are shown as necessary for explanation, and not all control lines and information lines are shown in the product. In practice, it can be considered that almost all configurations are interconnected.

1 Cluster division evaluation device 2 Network 3 Store server 4 Card company server 11 Central control device 12 Input device 13 Output device 14 Main memory device 15 Auxiliary storage device 16 Communication device 21 Survey value acquisition unit 22 Clustering unit 23 Regression analysis unit 24 Display processing Part 31 Prediction model 32, 32b Survey value information 33 Cluster information 34 Error information

Claims (9)

The survey value acquisition unit that acquires the survey value of the attribute of the legal entity and the survey value of the economic power of the trading entity,
A clustering unit that divides into a plurality of clusters in a multidimensional space based on the survey values of a plurality of attributes of the legal entity and the survey values of the economic power of the legal entity.
Based on the survey value of the attribute and the survey value of the economic power for each of the divided clusters , a prediction model is created with the attribute of the trading entity as the explanatory variable and the economic power of the trading entity as the objective variable, and the prediction is made. The model parameters are optimized for each of the plurality of clusters.
The difference between the predicted value of the economic power of the trading entity output by the prediction model optimized for the parameters and the survey value of the economic power of the trading entity is evaluated for each number of clusters.
A regression analysis unit that determines the number of clusters based on the evaluation results,
A cluster division evaluation device characterized by comprising. If the legal entity is a store
The attribute includes either the sales floor area, the property tax line price, or the number of parking lots.
If the legal entity is a customer
The attributes include either self-owned, annual income, payment method, or the purchase price of a particular item.
The cluster division evaluation device according to claim 1. The clustering unit
Accepting the user to select some or all of the plurality of explanatory variables.
The cluster division evaluation device according to claim 2. The regression analysis unit
Accepting the user to describe the mathematical formula of the prediction model on the screen, or displaying the model of the prediction model on the screen and accepting the user to select.
The cluster division evaluation device according to claim 3. Provided with a display processing unit that displays the number of the determined clusters and the evaluation result corresponding to the number of the clusters.
The cluster division evaluation device according to claim 1. The clustering unit
Accepting the user to enter the minimum and maximum values for the number of clusters and the minimum and maximum number of survey values contained in the clusters.
The cluster division evaluation device according to claim 5. If the legal entity is a store
The economic power is sales,
If the legal entity is a customer
The economic power is the purchase price,
The cluster division evaluation device according to claim 6. The survey value acquisition section of the cluster division evaluation device is
Obtain the survey value of the attribute of the trading entity and the survey value of the economic power of the trading entity,
The clustering unit of the cluster division evaluation device is
Divide into a plurality of clusters in a multidimensional space based on the survey values of a plurality of attributes of the legal entity and the survey values of the economic power of the legal entity.
The regression analysis unit of the cluster division evaluation device
Based on the survey value of the attribute and the survey value of the economic power for each of the divided clusters , a prediction model is created with the attribute of the trading entity as the explanatory variable and the economic power of the trading entity as the objective variable, and the prediction is made. The model parameters are optimized for each of the plurality of clusters.
The difference between the predicted value of the economic power of the trading entity output by the prediction model optimized for the parameters and the survey value of the economic power of the trading entity is evaluated for each number of clusters.
Determining the number of clusters based on the results of the evaluation,
A cluster division evaluation method for a cluster division evaluation device characterized by. Computer,
The survey value acquisition unit that acquires the survey value of the attribute of the legal entity and the survey value of the economic power of the trading entity,
A clustering unit that divides into a plurality of clusters in a multidimensional space based on the survey values of a plurality of attributes of the legal entity and the survey values of the economic power of the legal entity.
Based on the survey value of the attribute and the survey value of the economic power for each of the divided clusters , a prediction model is created with the attribute of the trading entity as the explanatory variable and the economic power of the trading entity as the objective variable, and the prediction is made. Create model parameters, optimize for each of the multiple clusters,
The difference between the predicted value of the economic power of the trading entity output by the prediction model optimized for the parameters and the survey value of the economic power of the trading entity is evaluated for each number of clusters.
A regression analysis unit that determines the number of clusters based on the evaluation results,
A cluster division evaluation program for functioning.

Images