Cancellation prediction system, cancellation prediction method, and cancellation prediction program

Abstract

This contract cancellation prediction system comprises an acquisition unit, a prediction unit, and an output unit. The acquisition unit acquires data that is of the situation of use of a communication line and that includes information concerning a contract of the communication line. The prediction unit predicts a contract cancellation probability of the communication line, on the basis of the acquired data of the situation of use by using a contract cancellation prediction model obtained by learning relationships between the situation of use of communication lines and the presence / absence of a contract cancellation of said communication lines. The output unit outputs the contract cancellation probability and a prediction reason by which the contract cancellation probability was predicted.

Description

[Technical Field]

[0001] The present invention relates to a churn prediction system. [Background technology]

[0002] When it comes to mobile phone contracts, switching to other carriers is a common practice. When taking actions aimed at maintaining a contract with a subscriber, using information on the cancellation probability of the contract can potentially enable effective actions.

[0003] The churn prediction system of Patent Document 1 classifies customers using their usage detail data, and displays a list of customers who are likely to churn based on the classification results and listing rules.

[0004] The customer retention support system of Patent Document 2 uses a wide variety of item data to extract customer groups for which a campaign to encourage the renewal of a service contract is highly effective.

[0005] The sales activity support device of Patent Document 3 uses customer information to predict the cancellation probability of customers who have already signed a contract, and generates a sales activity plan for non-customers whose tendencies match those of customers with a low cancellation probability.

[0006] The behavioral characteristic prediction system of Patent Document 4 extracts feature quantities that may affect user cancellation from a log of the communication status of a base station when a user communicates or makes a call. The behavioral characteristic prediction system of Patent Document 4 predicts user behavioral characteristics from the feature quantities extracted from the communication status log using a model that has learned the relationship between the extracted feature quantities and user behavioral characteristics. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Application Laid-Open No. 2002-334200 [Patent Document 2] Japanese Patent Application Laid-Open No. 2002-366732 [Patent Document 3] International Publication No. 2016 / 194628 [Patent Document 4] International Publication No. 2016 / 017086 Summary of the Invention [Problem to be solved by the invention]

[0008] However, with the techniques described in Patent Documents 1 to 4, it may be difficult to determine the validity of the prediction results.

[0009] An object of the present invention is to provide a churn prediction system etc. that makes it easy to judge the validity of the prediction results regarding the churn probability of a communication line. [Means for solving the problem]

[0010] In order to solve the above problems, the churn prediction system of the present invention comprises an acquisition means for acquiring data on communication line usage status including information on the communication line contract, a prediction means for predicting the churn probability of the communication line based on the acquired usage status data using a churn prediction model that has learned the relationship between the communication line usage status and whether or not the communication line has been churn based on the usage status, and an output means for outputting the churn probability and the reason for predicting the churn probability.

[0011] The churn prediction method of the present invention acquires data on the usage status of communication lines, including information on the contract for the communication lines, and predicts the churn probability of the communication lines based on the acquired usage status data using a churn prediction model that has learned the relationship between the usage status of the communication lines and whether or not the communication lines have been canceled, and outputs the churn probability and the reason for the predicted churn probability.

[0012] The program recording medium of the present invention records a churn prediction program that causes a computer to execute the following processes: a process of acquiring data on communication line usage status, including information on the communication line contract; a process of predicting the churn probability of a communication line based on the acquired usage status data, using a churn prediction model that has learned the relationship between the communication line usage status and whether or not the communication line has been canceled; and a process of outputting the churn probability and the reason for predicting the churn probability. [Effects of the Invention]

[0013] According to the present invention, it becomes easy to judge the validity of the predicted results regarding the cancellation probability of a communication line. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is a diagram illustrating an outline of the configuration of a first exemplary embodiment of the present invention. [Figure 2] 1 is a diagram illustrating an example of a configuration of a churn prediction system according to a first exemplary embodiment of the present invention. [Figure 3] FIG. 2 is a diagram illustrating an example of an operation flow of the churn prediction system according to the first exemplary embodiment of the present invention. [Figure 4] FIG. 2 is a diagram illustrating an example of a display screen according to the first embodiment of the present invention. [Figure 5] FIG. 2 is a diagram illustrating an example of a display screen according to the first embodiment of the present invention. [Figure 6] FIG. 2 is a diagram illustrating an example of an operation flow of the churn prediction system according to the first exemplary embodiment of the present invention. [Figure 7] FIG. 10 is a diagram illustrating an outline of the configuration of a second exemplary embodiment of the present invention. [Figure 8] FIG. 10 is a diagram illustrating an example of the configuration of a churn prediction system according to a second exemplary embodiment of the present invention. [Figure 9] FIG. 10 is a diagram illustrating an example of an operation flow of the churn prediction system according to the second exemplary embodiment of the present invention. [Figure 10] FIG. 10 is a diagram showing an example of a display screen according to the second embodiment of the present invention. [Figure 11] FIG. 10 is a diagram showing an example of a display screen according to the second embodiment of the present invention. [Figure 12]FIG. 10 is a diagram showing an example of a display screen according to the second embodiment of the present invention. [Figure 13] FIG. 10 is a diagram showing an example of a display screen according to the second embodiment of the present invention. [Figure 14] FIG. 10 is a diagram showing an example of a display screen according to the second embodiment of the present invention. [Figure 15] FIG. 10 is a diagram showing an example of a display screen according to the second embodiment of the present invention. [Figure 16] FIG. 10 is a diagram showing an example of a display screen according to the second embodiment of the present invention. [Figure 17] FIG. 10 is a diagram showing an example of a display screen according to the second embodiment of the present invention. [Figure 18] FIG. 10 is a diagram illustrating an example of the configuration of a churn prediction system according to a third exemplary embodiment of the present invention. [Figure 19] FIG. 10 is a diagram illustrating an example of an operation flow of the churn prediction system according to the third exemplary embodiment of the present invention. [Figure 20] FIG. 10 is a diagram illustrating an example of another configuration of an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0015] (First embodiment) A first embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 is a diagram showing an outline of the configuration of an information processing system of this embodiment. The information processing system of this embodiment includes a churn prediction system 10 and a terminal device 20. The churn prediction system 10 and the terminal device 20 are connected via a network. There may be multiple terminal devices 20. The number of terminal devices 20 can be set appropriately depending on the number of users of the churn prediction system 10.

[0016] The churn prediction system 10 is a system that predicts the churn probability of a communication line from the usage status of the communication line using a churn prediction model. The churn prediction model is generated by learning the relationship between the usage status of the communication line, including information about the communication line contract, and whether or not the communication line will be canceled. The information about the communication line contract is, for example, information about the contract details of the communication line and information about the terminal device used on the communication line. The contract details information of the communication line is, for example, conditions related to the contract period and fees.

[0017] A plurality of churn prediction models may be set. For example, the churn prediction system 10 performs churn prediction using different churn prediction models for churn prediction for a company that has contracts with multiple communication lines and for churn prediction for an individual. The churn prediction system 10 may predict the churn probability of a communication line using a different churn prediction model for each target for which churn prediction is performed. Data items of communication line usage status used for churn prediction are set for each churn prediction model.

[0018] The churn prediction model is generated as a learning model using a learning algorithm that can extract the prediction reason. An example of generating a learning model that can extract the prediction reason will be described later.

[0019] The usage status of a communication line may include, for example, data on one or more items of communication volume, number of communications, number of calls, call duration, and service or frequency of service use. The usage status of a communication line may also include information on the contract for the communication line. The information on the contract for the communication line may include, for example, items related to terminal information and contract content information. The terminal information may include, for example, data on one or more items of the terminal device's manufacturer, distributor, brand, model, color, form, specifications, additional functions, sales format, release date, release date of the next model, or frequency of model changes. The contract content information may include, for example, data on one or more items of subscriber information, contract plan, number of contracted lines, rate plan, contract period, contract period for long-term contracts, terminal device usage period, or frequency of past terminal replacement. If the subscriber is an individual, the subscriber information may include, for example, data on one or more items of the subscriber's age, gender, occupation, place of residence, or annual income. Furthermore, when the contractor is a company or organization that has contracts with multiple communication lines, data on one or more of the following items is used as contractor information: industry, business type, purpose of activity, performance, number of employees, number of bases, and years since establishment. Data on the usage status of communication lines, including information on the contracts for communication lines, may include data on items other than those mentioned above. Furthermore, each of the above items may be further divided into more detailed items.

[0020] The following describes the configuration of the churn prediction system 10. Figure 2 is a diagram showing an example of the configuration of the churn prediction system 10. The churn prediction system 10 includes an acquisition unit 11, a prediction unit 12, an output unit 13, a prediction model generation unit 14, and a storage unit 15.

[0021] The acquisition unit 11 acquires data on the usage status of the communication line, including information on the contract for the communication line. Furthermore, when multiple types of churn prediction models are used in the prediction unit 12, the acquisition unit 11 may acquire a selection result of a churn prediction model to be used for churn prediction. That is, the selection result of the churn prediction model to be used for churn prediction may be, for example, a selection result of a target of churn prediction. In the case of the selection result of a target of churn prediction, a correspondence between the target of churn prediction and the churn prediction model to be used for the target of churn prediction is set in advance. Furthermore, the acquisition unit 11 may acquire data on the usage status of the communication line and data on whether the communication line has been canceled. The data on the usage status of the communication line and the data on whether the communication line has been canceled are used to generate a churn prediction model.

[0022] The acquisition unit 11 acquires data on the usage status of the communication line, data on whether the communication line has been canceled, and the selection result of the churn prediction model from the terminal device 20. The data on the usage status of the communication line, data on whether the communication line has been canceled, and the selection result of the churn prediction model are input to the terminal device 20 by, for example, an operator's operation. The acquisition unit 11 may acquire data on the usage status of the communication line and data on whether the communication line has been canceled from a contract management server connected via a network. Furthermore, the data on the usage status of the communication line, data on whether the communication line has been canceled, and the selection result of the churn prediction model may be input to the churn prediction system 10 by an operator's operation.

[0023] The churn prediction model may be one type regardless of the target of churn prediction. Also, the churn prediction model used for churn prediction may be set for each terminal device 20 or user of the terminal device 20 accessing the churn prediction system 10. If there is one type regardless of the target of churn prediction, the acquisition unit 11 does not need to acquire the selection result of the churn prediction model.

[0024] The prediction unit 12 predicts the churn probability of a communication line based on data on the usage status of the communication line, using a churn prediction model that has learned the relationship between the usage status of the communication line and whether or not the communication line will be canceled. The churn prediction model may be a model generated by the prediction model generation unit 14, or may be a model generated outside the churn prediction system 10. The churn prediction model is a model generated using a learning algorithm that is capable of outputting the predicted reason for the churn probability as a prediction result.

[0025] The prediction unit 12 may predict the churn probability of a communication line using one type of churn prediction model, or may predict the churn probability of a communication line by selecting and using any one or more churn prediction models from multiple types of churn prediction models. When multiple types of churn prediction models are used, the prediction unit 12 predicts the churn probability of a communication line using, for example, a churn prediction model specified by the selection result of the churn prediction model acquired by the acquisition unit 11.

[0026] Furthermore, the prediction unit 12 predicts the churn probability from data on the usage status of communication lines, for example, using either a churn prediction model for a group subscribing to multiple communication lines, or a churn prediction model for a communication line unit. Three or more churn prediction models may be used. For example, when a churn prediction model for a group is selected as the prediction target, the prediction unit 12 predicts the churn probability for each communication line from the usage status of each of multiple communication lines, including communication lines subscribed to by the same subscriber, using the selected churn prediction model.

[0027] When one type of churn prediction model is used regardless of the target of churn prediction, the prediction unit 12 predicts the churn probability using that churn prediction model. When a churn prediction model is set according to the terminal device 20 or the user of the terminal device 20, the prediction unit 12 predicts the churn probability using the set churn model.

[0028] The prediction unit 12 performs churn prediction by predicting the churn probability of the communication line using a churn prediction model, with communication line usage status data including information about the communication line contract as input data. Of the communication line usage status data, data items used to predict the churn probability of the communication line are set for each churn prediction model. The prediction unit 12 uses, as input data, information about the communication line contract, such as at least one of the usage period of the terminal device used in the contract, the period since contract renewal, or the period since contract initiation. For example, the contract period of a long-term contract and the usage period of a terminal device for which installment payments are made may have a significant impact on whether or not the contract is renewed. Therefore, the prediction unit 12 can improve the accuracy of predicting the churn probability by using information about the contract and the period related to the terminal device from the information about the communication line contract.

[0029] The prediction unit 12 extracts prediction reasons for the churn probability when predicting the churn probability of a communication line. The prediction unit 12 extracts, from the data on the usage status of the communication line, items whose influence on the churn probability of the communication line meets a criterion as prediction reasons. For example, the prediction unit 12 extracts, as prediction reasons, items whose influence on the churn probability is greater than other items. The prediction unit 12 may extract multiple items as prediction reasons. The prediction unit 12 may extract, as prediction reasons, a predetermined number of items in descending order of their influence on the value of the churn probability. Furthermore, the prediction unit 12 may extract, as prediction reasons, items whose influence on the value of the churn probability is equal to or greater than a predetermined criterion. The criterion for extracting prediction reasons may be set as appropriate.

[0030] The output unit 13 outputs the churn probability and the prediction reason for predicting the churn probability. The output unit 13 outputs the churn probability predicted by the prediction unit 12 using the churn prediction model and the prediction reason for the churn probability as prediction results, for example to the terminal device 20. The output unit 13 outputs, as the prediction reason, items of the communication line usage status data whose influence on the churn probability meets a criterion. The output unit 13 may output data displaying the churn probability and the prediction reason to a display device (not shown) connected to the churn prediction system 10.

[0031] The prediction model generation unit 14 generates a churn prediction model by learning the relationship between the usage status of the communication line and whether or not the communication line will be canceled. When there are multiple types of churn prediction models, the prediction model generation unit 14 may generate a churn prediction model for each target of churn prediction, for example. The prediction model generation unit 14 stores the generated churn prediction model in the storage unit 15.

[0032] The prediction model generation unit 14 generates a churn prediction model using a learning algorithm that can output the predicted reason for the churn probability as a prediction result.

[0033] The prediction model generation unit 14 generates a churn prediction model using, for example, a learning algorithm based on factorized asymptotic Bayesian inference. When performing learning using a learning algorithm based on factorized asymptotic Bayesian inference, the prediction model generation unit 14 uses communication line usage status data as input data and the presence or absence of churn as correct answer data, performs case classification using decision tree rules, and generates a learning model that predicts churn probability using a linear model that combines different explanatory variables for each case. The prediction model generation unit 14 generates the learning model by sequentially optimizing the classification conditions for the data, generating a prediction model by optimizing the combination of explanatory variables, and deleting unnecessary prediction models. This method of generating a learning model is called heterogeneous mixture learning because it combines prediction models with different combinations of explanatory variables. Generating a churn prediction model using heterogeneous mixture learning makes it possible to explain the predicted churn probability using case classification conditions that have a strong impact on the prediction result, thereby improving the interpretability of the prediction result. A heterogeneous mixture learning technique is disclosed, for example, in U.S. Patent Application Publication No. 2014 / 0222741.

[0034] The learning algorithm used to generate the churn prediction model is not limited to the above example. For example, the prediction model generation unit 14 may generate a learning model that predicts churn probability from data on communication line usage status by deep learning using a neural network. When using such a learning model, the prediction model generation unit 14 generates a churn prediction model that outputs predicted reasons for churn probability by, for example, varying the data for each item and extracting items that have a large impact on churn probability as prediction reasons based on the amount of change in the value of churn probability.

[0035] The prediction model generation unit 14 verifies the accuracy of the generated churn prediction model when determining whether the generation of the churn prediction model has been completed. The prediction model generation unit 14 uses, for example, verification data as input data and predicts the churn probability using the churn prediction model being generated. The verification data may be, for example, data on the usage status of communication lines and data on whether or not a churn has occurred that has been acquired by the acquisition unit 11 and that has not been used to generate the churn prediction model.

[0036] The prediction model generation unit 14 determines that the result of the churn prediction is correct when the churn probability is equal to or higher than a preset standard and the data indicating whether or not churn occurs indicates churn. The prediction model generation unit 14 predicts the churn probability using, for example, verification data as input data, and determines that the generation of the churn prediction model is complete when the accuracy rate is equal to or higher than the standard. When the prediction model generation unit 14 determines that the generation of the churn prediction model is complete, it stores the churn prediction model in the storage unit 15 as a trained learning model.

[0037] The prediction model generation unit 14 may update the churn prediction model by re-learning the relationship between the usage status of the communication line for which the prediction was made and the actual data on whether or not there was a churn. When re-learning is performed, the prediction model generation unit 14 updates the churn prediction model stored in the storage unit 15 using each parameter after the re-learning.

[0038] The storage unit 15 stores each data related to churn prediction. The storage unit 15 stores the data on the usage status of the communication line, the data on whether or not there is a churn, and the selection result of the churn prediction model acquired by the acquisition unit 11. The storage unit 15 stores the churn probability predicted by the prediction unit 12 and the predicted reason for the churn probability as prediction results.

[0039] The storage unit 15 stores the churn prediction model generated by the prediction model generation unit 14. When multiple churn prediction models are generated, the storage unit 15 may store each churn model. Furthermore, when a churn prediction model is generated for each churn prediction target, the storage unit 15 may store the churn prediction model in association with the churn prediction target.

[0040] The terminal device 20 displays the churn probability of the communication line and the predicted reason for the churn probability, which are obtained as prediction results from the churn prediction system 10, on a display device (not shown). When making a churn prediction, the terminal device 20 accepts communication line usage status data and a selection of a churn prediction model input by an operator's operation. When generating a churn prediction model, the terminal device 20 also accepts communication line usage status data and data on whether the communication line has been churn, input by an operator's operation. When making a churn prediction prediction model, the terminal device 20 outputs the communication line usage status data and the selection result of the churn prediction model to the churn prediction system 10. When generating a churn prediction model, the terminal device 20 also outputs communication line usage status data and data on whether the communication line has been churn.

[0041] Input of communication line usage status data when predicting churn, and input of communication line usage status data and churn / non-churn data when generating a churn prediction model may be performed on different terminal devices 20. Furthermore, input of communication line usage status data when predicting churn, and display of the churn probability of the communication line and the prediction reason, which are the prediction results, may be performed on different terminal devices 20.

[0042] For example, a personal computer is used as the terminal device 20. Furthermore, the terminal device 20 for displaying the prediction results may be, for example, a tablet computer or a smartphone. The terminal device 20 is not limited to these examples.

[0043] The operation of the information processing system of this embodiment when predicting churn of a communication line will be described below. Fig. 3 is a diagram showing an example of the operation flow of the churn prediction system 10 when predicting churn.

[0044] The acquisition unit 11 of the churn prediction system 10 acquires data on the usage status of the communication line that is the target of churn prediction from the terminal device 20 (step S11). The data on the usage status of the communication line is input to the terminal device 20 by, for example, an operator's operation.

[0045] Furthermore, the acquisition unit 11 acquires the selection result of the churn prediction model from the terminal device 20 (step S12). The selection result of the churn prediction model is input to the terminal device 20 by, for example, an operation by an operator. When only one churn prediction model is determined, the acquisition unit 11 does not need to acquire the selection result of the churn prediction model. In such a case, the prediction unit 12 performs churn prediction using the only one churn prediction model.

[0046] When the data on the usage status of the communication line and the selection result of the churn prediction model are acquired, the prediction unit 12 reads out the churn prediction model corresponding to the selection result from the storage unit 15. After reading out the churn prediction, the prediction unit 12 predicts the churn probability of the communication line using the read-out churn prediction model, with the data on the usage status of the communication line as input data (step S13).

[0047] When predicting the churn probability, the prediction unit 12 extracts, from among the items of the data on the usage status of the communication line used as input data, items that have a large impact on the churn probability as prediction reasons. A large impact on the churn probability means, for example, that when the value of that item fluctuates, the change in the value of the churn probability is larger than that of other items.

[0048] After the churn probability is predicted and the reason for the prediction is extracted, the output unit 13 outputs the churn probability of the communication line and the predicted reason for the churn probability to the terminal device 20 (step S14).

[0049] When the churn probability of the communication line and the predicted reason for the churn probability are acquired, the terminal device 20 displays the churn probability of the communication line and the predicted reason for the churn probability on the display device.

[0050] FIG. 4 is a diagram showing an example of a display screen showing the predicted results of the churn probability for each subscriber when churn prediction is performed using a churn prediction model that is applied when the subscriber of a communication line is an individual.

[0051] In the example of Fig. 4, the output unit 13 outputs a list of "subscribers" indicating subscribers of the communication line, "churn probabilities" indicating the churn probability of the communication line, and "reasons" indicating the predicted reasons as a churn possibility list. The terminal device 20 displays a display screen as shown in Fig. 4 based on the acquired data. In the example of Fig. 4, the output unit 13 outputs a churn possibility list indicating the churn probabilities and predicted reasons for subscribers in descending order of churn probability as a prediction result.

[0052] Fig. 5 is a diagram showing another example of a display screen showing the churn probability of each communication line subscribed by a subscriber and the predicted reasons for the churn probability. Fig. 5 shows an example in which a churn prediction model is used when predicting churn for a subscriber who subscribes to multiple communication lines. In the example of Fig. 5, the output unit 13 outputs a list of "telephone numbers" indicating the telephone numbers assigned to the communication lines, "churn probabilities" indicating the churn probability of the communication lines, and "reasons" indicating the predicted reasons for the churn probability, as a churn probability list for the communication lines subscribed to by "Company M."

[0053] Examples of the display screen based on the prediction result output by the output unit 13 are not limited to the examples in Fig. 4 and Fig. 5. The output unit 13 may output, for example, the average churn probability for communication lines subscribed to by each subscriber or the number of communication lines with a churn probability equal to or higher than a standard as the prediction result. In such a configuration, the output unit 13 may output the prediction result as a list of the average churn probability for each company or the number of communication lines with a churn probability equal to or higher than a standard.

[0054] The operation of generating a churn prediction model will now be described. Fig. 6 is a diagram showing an example of the operation flow of the churn prediction system 10 when generating a churn prediction model.

[0055] The acquisition unit 11 of the churn prediction system 10 acquires data on the usage status of communication lines and data on whether or not a communication line has been canceled, which data are used to generate a churn prediction model (step S21).

[0056] When the data on the usage status of the communication line and the data on whether or not the contract has been cancelled are acquired, the prediction model generation unit 14 generates a churn prediction model by learning the relationship between the data on the usage status of the communication line and whether or not the contract has been cancelled (step S22). The prediction model generation unit 14 generates the churn prediction model by, for example, heterogeneous mixture learning.

[0057] Once the churn prediction model has been generated, the prediction model generation unit 14 determines whether or not the generation of the churn prediction model has been completed. When determining whether or not the generation of the churn prediction model has been completed, the prediction model generation unit 14 verifies the accuracy of the predictions of the churn prediction model. The prediction model generation unit 14 uses, for example, verification data as input data and calculates the churn probability using the churn prediction model. The verification data is data that has not been used to generate the churn prediction model, out of the communication line usage status data and churn data acquired by the acquisition unit 11.

[0058] If the churn probability value predicted using the churn prediction model is equal to or greater than a preset standard and the churn presence / absence data indicates that the communication line has been canceled, the prediction model generation unit 14 determines that the churn prediction result is correct. The prediction model generation unit 14 determines whether the termination condition for generating the churn prediction model is met depending on whether the accuracy rate of the churn prediction result meets the standard. If the accuracy of the churn prediction model is equal to or greater than the standard, that is, if the accuracy rate of the churn prediction result is equal to or greater than the standard (Yes in step S23), the prediction model generation unit 14 stores the churn prediction model in the storage unit 15 as a trained learning model (step S24). If the accuracy of the churn prediction model is less than the standard (No in step S23), the prediction model generation unit 14 returns to step S22 and continues generating the churn prediction model.

[0059] The churn prediction system 10 of this embodiment extracts the churn probability of a communication line as well as the predicted reason for the churn probability, and outputs the churn probability and the predicted reason for the churn probability as prediction results. By outputting the churn probability of a communication line and the predicted reason, for example, when taking some action against a subscriber, it is possible to determine the validity of the predicted result of the churn probability of a communication line and take action according to the predicted reason. As a result, by using the churn prediction system 10 of this embodiment, it becomes easy to determine the validity of the predicted result of the churn probability of a communication line.

[0060] Furthermore, the churn prediction system 10 of this embodiment predicts the churn probability of a communication line using a churn prediction model appropriate for the prediction target. For example, the churn prediction system 10 makes churn predictions using different churn prediction models depending on whether the prediction target is an individual or a company or organization that has contracts with multiple lines. In this way, when predictions are made using a churn prediction model appropriate for the prediction target, the churn prediction system 10 can improve the accuracy of churn predictions.

[0061] (Second embodiment) A second embodiment of the present invention will be described in detail with reference to the drawings. FIG. 7 is a diagram showing an outline of the configuration of an information processing system of this embodiment. The information processing system of this embodiment includes a churn prediction system 30 and a terminal device 40. The churn prediction system 30 and the terminal device 40 are connected via a network. There may be multiple terminal devices 40. The number of terminal devices 40 can be set appropriately depending on the number of users of the churn prediction system 30.

[0062] The churn prediction system 10 of the first embodiment outputs the churn probability of a communication line and the prediction result of the churn probability. In contrast, the churn prediction system 30 of the present embodiment acquires the selection result of any one of the display items of the output prediction result, and further outputs data on the selected item.

[0063] The configuration of the churn prediction system 30 will be described. Fig. 8 is a diagram showing an example of the configuration of the churn prediction system 30. The churn prediction system 30 includes an acquisition unit 31, a prediction unit 12, an output unit 32, a prediction model generation unit 14, and a storage unit 15. In the churn prediction system 30, the configurations and functions of the prediction unit 12, the prediction model generation unit 14, and the storage unit 15 are similar to the components of the same names in the churn prediction model in the first embodiment.

[0064] The acquisition unit 31 has the same functions as the acquisition unit 11 of the first embodiment. In addition to the functions of the acquisition unit 11, the acquisition unit 31 acquires, as a selection result, an item selected on a display screen of the prediction result of the churn prediction on the terminal device 40. The item selected on the display screen of the prediction result indicates an item to be displayed in more detail on the display screen.

[0065] The acquisition unit 31 acquires, from the terminal device 40, data on the usage status of the communication line, data on whether the communication line contract has been canceled, the selection result of the churn prediction model, and the items selected on the display screen as selection results. The data on the usage status of the communication line, data on whether the communication line contract has been canceled, the selection result of the churn prediction model, and the selection result of the items selected on the display screen are input to the terminal device 40 by, for example, operation by an operator. Furthermore, the selection results of the display items are input to the terminal device 40 by operation by an operator. The selection results of the display items may also be input to the churn prediction system 30 by operation by an operator.

[0066] The output unit 32 has the same functions as the output unit 13 of the first embodiment. In addition to the functions of the output unit 13, the output unit 32 outputs data according to the selection result of the display items acquired via the acquisition unit 31 to the terminal device 40. The selection result of the display items is data indicating the items selected by the operator's operation on the output screen of the prediction result on the terminal device 40.

[0067] The terminal device 40 has the same functions as the terminal device 20 of the first embodiment. In addition to the functions of the terminal device 20, the terminal device 40 displays display data related to an item selected on the display screen. The terminal device 40 displays the display data related to an item selected on the display screen on the display device based on data acquired from the churn prediction system 30. The terminal device 40 accepts the selection of any item on the display screen as a selection result, which is input by the operator's operation. The terminal device 40 transmits the item selection result to the churn prediction system 30.

[0068] For example, a personal computer is used as the terminal device 40. Furthermore, the terminal device 40 for displaying the prediction results may be, for example, a tablet computer or a smartphone. However, the terminal device 40 is not limited to these examples.

[0069] The operation of the information processing system of this embodiment when predicting churn of a communication line will be described below. Fig. 9 is a diagram showing an example of the operation flow of the churn prediction system 30 when predicting churn.

[0070] The acquisition unit 31 of the churn prediction system 30 acquires data on the usage status of the communication line that is the target of churn prediction from the terminal device 40 (step S31). The data on the usage status of the communication line is input to the terminal device 40 by, for example, an operator's operation.

[0071] Furthermore, the acquisition unit 31 acquires the selection result of the churn prediction model from the terminal device 40 (step S32). The selection result of the churn prediction model is input to the terminal device 40 by, for example, an operation by an operator.

[0072] When the communication line usage status data and the selection result of the churn prediction model are acquired, the prediction unit 12 reads out the churn prediction model corresponding to the selection result from the storage unit 15. After reading out the churn prediction, the prediction unit 12 predicts the churn probability of the communication line using the read churn prediction model with the communication line usage status data as input data (step S33). Furthermore, the prediction unit 12 extracts, from among the items of the communication line usage status data used as input data, items that have a large impact on the churn probability as prediction reasons.

[0073] After the churn probability is predicted and the predicted reason is extracted, the output unit 32 outputs the churn probability of the communication line and the predicted reason for churn to the terminal device 40 (step S34). Furthermore, the terminal device 40 adds information on items for which more detailed data can be output from among the output data to the churn probability of the communication line and the predicted reason for churn, and outputs the information to the terminal device 40.

[0074] When the churn probability of the communication line and the predicted reason for the churn probability are acquired, the terminal device 40 displays the churn probability of the communication line and the predicted reason for the churn probability on the display device. Furthermore, the terminal device 40 displays items for which more detailed data can be displayed in a manner that allows the item to be selected. "Selectable items" means that the worker can select an item by clicking it with a mouse, for example. The method for selecting an item is not limited to clicking it with a mouse, and can be set as appropriate.

[0075] Fig. 10 is a diagram showing an example of a display screen of the predicted churn probability for each subscriber when churn prediction is performed using a churn prediction model applied when the subscriber of a communication line is an individual. In the example of Fig. 10, as in Fig. 4, a list of "subscriber" indicating the subscriber of the communication line, "churn probability" indicating the churn probability of the communication line, and "reason" indicating the predicted reason is displayed as a churn possibility list. In the example of Fig. 10, the churn probability and predicted reasons are displayed in descending order of subscriber churn probability.

[0076] In the example of FIG. 10, each piece of data indicated in the reason can be selected by clicking it with a mouse, for example. Also, in the example of FIG. 10, the bottom row has selection buttons for "New Prediction" and "End." "New Prediction" is a button that is selected when, for example, changing the prediction target or the churn prediction model and making a churn prediction. Also, "End" is a button that is selected when ending the churn prediction. These setting buttons can be set according to the usage form of the churn prediction system 30. Also, the functions of the selection buttons for "New Prediction" and "End" in the example display screens below are the same as those in FIG. 10.

[0077] In the processing of step S34, when the screen showing the churn probability and the predicted reason for the churn probability is displayed, if the operator selects one of the items selectable on the display screen, the terminal device 40 outputs the data of the selected item as the selection result to the churn prediction system 30.

[0078] In this case, the acquisition unit 31 acquires the selection result of the screen to be displayed from the terminal device 40. When the selection result of the screen to be displayed is acquired (Yes in step S35), the output unit 32 outputs data according to the selection result to the terminal device 40 (step S36).

[0079] When the data corresponding to the selection result is acquired, the terminal device 40 uses the acquired data to display a display screen corresponding to the selection result on the display device.

[0080] Fig. 11 shows an example of a display screen when "Terminal Usage Period" is selected in the example of Fig. 10. In the example of Fig. 11, subscribers for whom "Terminal Usage Period" has the greatest influence on the prediction result are displayed in a list in descending order of churn probability. In the example of Fig. 11, the output unit 32 outputs the churn probability for communication lines for which the probability of churn due to the selected prediction reason "Terminal Usage Period" is equal to or higher than a standard. In the example of FIG. 10, when the selection result in which "terminal usage period" is selected is acquired via the acquisition unit 31, the output unit 32 generates data of a list of subscribers whose "terminal usage period" has the greatest impact on the prediction result, and outputs it to the terminal device 40. The terminal device 40 uses the data acquired from the churn prediction system 30 to display on the display device a list of subscribers whose "terminal usage period" has the greatest impact on the prediction result, as shown by way of example in FIG. 11. The "Back" selection button shown in the lower part of FIG. 11 is selected, for example, when returning from the display screen shown in FIG. 11 to the display screen shown in FIG. 10.

[0081] After outputting data according to the selection result to the terminal device 40, when an instruction to end churn prediction is received (Yes in step S37), the churn prediction system 30 ends the operation related to churn prediction. The instruction to end churn prediction is output from the terminal device 40 to the churn prediction system 30 when, for example, the operator selects the "End" button in Figure 10.

[0082] Furthermore, even if no output data selection result is input (No in step S35) and an instruction to end churn prediction is received (Yes in step S37), the churn prediction system 30 ends the operation related to churn prediction.

[0083] If no instruction to end churn prediction is input in step S37 (No in step S37), the churn prediction system returns to step S34 and waits until a selection result on the display screen is input.

[0084] The method of generating a churn prediction model in the churn prediction system 30 is the same as the method of generating a churn prediction model in the churn prediction system 10 of the first embodiment.

[0085] An example of the output of the churn probability and the reason for the churn probability by the output unit 32 will be further described. The output unit 32 may output, for example, an aggregated value based on the churn probability of each communication line for each subscriber as the prediction result. For example, the aggregated value for each subscriber is the average churn probability of the communication lines subscribed to by the subscriber, or the number of communication lines with a churn probability equal to or higher than a standard. The aggregated value for each subscriber is not limited to the above example.

[0086] Fig. 12 shows an example of a display screen that displays a list of subscribers with a high average churn probability for the lines they have subscribed to when performing churn prediction for subscribers who have subscribed to multiple communication lines. In the example of Fig. 12, a list of "contracted companies" indicating the names of companies that have subscribed to communication lines and "average churn probability" indicating the average churn probability for the lines that each company has subscribed to is displayed as a churn possibility list. In the example of Fig. 12, the list is generated in descending order of the average churn probability of the companies.

[0087] 12, each of the contracted company names can be selected by clicking it with a mouse, for example. When the operator clicks on one of the company names, the terminal device 40 outputs the clicked company name to the churn prediction system 30 as the selection result.

[0088] Fig. 13 is a diagram showing an example of a display screen that displays a list of the churn probability for each line subscribed to by the selected company when any of the companies is selected in the example of Fig. 12. The example of Fig. 13 displays the "telephone number" of each line subscribed to by the company selected in Fig. 12, "churn probability" showing the predicted churn probability for each line, and "churn factors" showing data on items that have a large impact on the churn probability. In the example of Fig. 13, the "churn factors" displayed are "contract period," "period since renewal," and "terminal usage period."

[0089] 12 is displayed, and if a selection result in which "terminal usage period" is selected is acquired via the acquisition unit 31, the output unit 32 outputs data on the subscriber for whom "terminal usage period" has the greatest impact on the prediction result to the terminal device 40. Using the data acquired from the churn prediction system 30, the terminal device 40 displays on the display device a display screen listing the churn probabilities of each line for the selected company, as shown in the example of FIG.

[0090] FIG. 14 is a diagram illustrating an example of a display screen displaying a list of predicted churn prediction results for each communication line and data for each item. In the example of FIG. 14, "Subscriber" indicating the name of the subscriber for each communication line, "Churn Probability" indicating the predicted churn probability for each line, and "Churn Factors" indicating data on items that have a large impact on the churn probability are displayed. In the example of FIG. 14, "Contract Period," "Period from Renewal," and "Terminal Usage Period" are displayed as "Churn Factors." The display screen illustrated in the example of FIG. 14 is displayed, for example, when a button for selecting "Detailed Display" is set on the display screen illustrated in the example of FIG. 10. Furthermore, the output unit 32 may output the data for the display screen illustrated in the example of FIG. 14 as the prediction result to the terminal device 40, instead of the data for the display screen illustrated in the example of FIG. 10.

[0091] FIG. 15 shows an example of a display screen in which items that have a large impact on churn probability are highlighted. FIG. 15 shows an example of a display screen similar to the example of FIG. 14 , in which items that have a large impact on churn probability are highlighted. When displaying the display screen shown in FIG. 15 , the output unit 32 adds information indicating a large impact on churn probability to data on items that have a large impact on churn probability among data on each item of churn factors, and outputs the added data to the terminal device 40. The terminal device 40 displays a display screen on the display device in which the items to which information indicating a large impact on churn probability is added are highlighted as shown in FIG. 15 . The highlighting of items that have a large impact on churn probability can be achieved, for example, by using a different color within the display frame, thickness of the display frame, color of the text, size of the text, thickness of the text, or font of the text from other items. Furthermore, the highlighting of items that have a large impact on churn probability may be achieved by a combination of multiple methods. Furthermore, the method of highlighting items that have a large impact on churn probability is not limited to the above example.

[0092] Fig. 16 is a diagram showing an example of a display screen that shows factors for contract renewal along with factors for contract cancellation. In the example of Fig. 16, "Contractor" showing the name of the subscriber of each communication line, "Cancelation Probability" showing the predicted result of the churn probability for each line, "Cancelation Factors" showing items that have a large impact on the predicted result, and "Continuation Factors" showing items that are factors for contract renewal of the communication line are displayed.

[0093] The items displayed as continuation factors are set in advance by, for example, analyzing data on communication lines whose contracts were renewed among contracts with similar trends in cancellation factors, and are saved as a data table. In the data table showing continuation factors, for example, for each cancellation factor item, the continuation factor is set so that it corresponds to the criteria for other items. The output unit 32 extracts from the data table the continuation factors that correspond to the cancellation factors and usage status data for the communication line for which the prediction results are to be output.

[0094] If the churn prediction model is generated using a learning algorithm that can predict not only the churn probability but also the probability of not churn, i.e., the continuation probability, the output unit 32 may output items that have a large impact on the continuation probability as continuation factors. The output unit 32 associates the extracted data on continuation factors with the churn probability of the communication line and information on the churn factors, and outputs the data to the terminal device 40.

[0095] The output unit 32 may output recommended actions for a contract whose churn probability meets a standard in association with the churn probability and the predicted reason. Fig. 17 shows an example of a display screen that displays recommended actions for a contractant whose churn probability meets a standard. For example, the output unit 32 outputs recommended actions for a contract whose churn probability is equal to or higher than a standard in association with the churn probability and churn factors of the communication line.

[0096] The example in Figure 17 shows "Subscriber," which indicates the name of the subscriber for each communication line, "Canceler Probability," which indicates the predicted result of the churn probability for each line, "Canceler Causes," which indicates the items that have a large impact on the predicted result, and "Recommended Actions," which indicates the recommended actions to continue the service.

[0097] The items displayed as recommended actions are preset based on, for example, actions taken in the past for subscribers of communication lines with similar trends in cancellation factors, which resulted in contract renewals. The set items are saved as a data table. When outputting recommended actions, the output unit 32 searches the data table and extracts data on recommended actions corresponding to the communication line for which the prediction results are to be output. After extracting the recommended actions, the output unit 32 associates the extracted recommended actions with the cancellation probability and cancellation factors of the communication line and outputs them to the terminal device 40.

[0098] Examples of the display screens displayed by the churn prediction system 30 outputting data to the terminal device 40 are not limited to the examples shown above. In addition, the configurations of the above display screens may be combined as appropriate.

[0099] The churn prediction system 30 of the information processing system of this embodiment outputs data corresponding to the results selected on the churn prediction output screen to the terminal device 40. With this configuration, the churn prediction system 30 can, for example, output only the churn probability and the prediction reason as a list, and output more detailed items upon request, thereby enabling both an understanding of the overall results and confirmation of detailed data. Furthermore, by switching the displayed items and referring to each piece of information, the user of the churn prediction system 30 can appropriately understand the churn probability and churn factors of the communication line and take action to prevent churn. Furthermore, by enabling both an understanding of the overall results and confirmation of detailed data, the churn prediction system 30 can easily determine the validity of the prediction results for the churn probability of the communication line.

[0100] (Third embodiment) The third embodiment of the present invention will be described in detail with reference to the drawings. Fig. 18 is a diagram showing an example of the configuration of a churn prediction system 100 of this embodiment.

[0101] The churn prediction system 100 includes an acquisition unit 101, a prediction unit 102, and an output unit 103. The acquisition unit 101 acquires data on the usage status of communication lines, including information on the contract for the communication lines. The prediction unit 102 predicts the churn probability of the communication line based on the acquired usage status data, using a churn prediction model that has learned the relationship between the usage status of the communication line and whether or not the communication line will be canceled. The output unit 103 outputs the churn probability and the reason for predicting the churn probability.

[0102] The acquisition unit 11 of the first embodiment and the acquisition unit 31 of the second embodiment are examples of the acquisition unit 101. The acquisition unit 101 is also an aspect of acquisition means. The prediction unit 12 of the first embodiment and the second embodiment is an example of the prediction unit 102. The prediction unit 102 is an aspect of prediction means. The output unit 13 of the first embodiment and the output unit 32 of the second embodiment are examples of the acquisition unit 101. The output unit 103 is also an aspect of output means.

[0103] A description will be given of the operation of the churn prediction system 100. Fig. 19 is a diagram showing an example of the operation flow of the churn prediction system 100.

[0104] The acquisition unit 101 acquires data on the usage status of a communication line, including information on the contract for the communication line (step S101). Once the usage status data has been acquired, the prediction unit 102 predicts the churn probability of the communication line based on the acquired usage status data, using a churn prediction model that has learned the relationship between the usage status of the communication line and whether or not the communication line has been churn (step S102). Once the churn probability of the communication line has been predicted, the output unit 103 outputs the churn probability and the reason for predicting the churn probability (step S103).

[0105] The churn prediction system 100 of this embodiment predicts the churn probability of a communication line using a churn prediction model from data on the usage status of the communication line, including information on the contract for the communication line. The churn prediction system 100 also outputs the predicted churn probability and the reason for predicting the churn probability. By using the churn prediction system 100, the churn probability of a communication line can be obtained together with the reason for the prediction, making it easier to determine the validity of the prediction result for the churn probability of the communication line.

[0106] Each process in the churn prediction system 10 of the first embodiment, the churn prediction system 30 of the second embodiment, and the churn prediction system 100 of the third embodiment can be realized by executing a computer program on a computer. Fig. 20 shows an example of the configuration of a computer 200 that executes a computer program that performs each process in the churn prediction system 10 of the first embodiment, the churn prediction system 30 of the second embodiment, and the churn prediction system 100. The computer 200 includes a CPU (Central Processing Unit) 201, a memory 202, a storage device 203, an input / output I / F (Interface) 204, and a communication I / F 205.

[0107] The CPU 201 reads and executes computer programs for performing each process from the storage device 203. The CPU 201 may be configured by a combination of multiple CPUs. The CPU 201 may also be configured by a combination of a CPU and another type of processor. For example, the CPU 201 may be configured by a combination of a CPU and a graphics processing unit (GPU). The memory 202 is configured by a dynamic random access memory (DRAM) or the like, and temporarily stores computer programs executed by the CPU 201 and data being processed. The storage device 203 stores computer programs executed by the CPU 201. The storage device 203 is configured by, for example, a non-volatile semiconductor storage device. Other storage devices such as a hard disk drive may also be used for the storage device 203. The input / output I / F 204 is an interface that receives input from an operator and outputs display data, etc. The communication I / F 205 is an interface that transmits and receives data to and from a terminal device or another information processing device. The terminal device 20 of the first embodiment and the terminal device 40 of the second embodiment may also have a similar configuration.

[0108] The processes in the churn prediction system 10 of the first embodiment, the churn prediction system 30 of the second embodiment, and the churn prediction system 100 of the third embodiment may be performed on multiple computers connected via a network. For example, the process related to churn prediction and the process related to generating a churn prediction model may be performed on different computers. Furthermore, the generation of a churn prediction model and the generation of a churn prediction model by relearning may be performed on different computers. Furthermore, the memory units of the churn prediction system 10 and the churn prediction system 30 may be provided in a storage device connected via a network or a storage device managed by a server connected via a network.

[0109] The computer program used to execute each process can also be stored on a recording medium and distributed. Examples of recording media that can be used include magnetic tapes for recording data and magnetic disks such as hard disks. Optical disks such as CD-ROMs (Compact Disc Read Only Memory) can also be used as recording media. Non-volatile semiconductor storage devices can also be used as recording media.

[0110] A part or all of the above-described embodiments can be described as, but not limited to, the following supplementary notes.

[0111] [Appendix 1] an acquisition means for acquiring data on the usage status of the communication line, including information on the contract for the communication line; a prediction means for predicting a churn probability of the communication line based on the acquired data on the usage status, using a churn prediction model that has learned the relationship between the usage status of the communication line and whether or not the communication line is churn; an output means for outputting the churn probability and a prediction reason for predicting the churn probability; A cancellation prediction system equipped with

[0112] [Appendix 2] The output means outputs the prediction reason using an item of the usage status data whose influence on the cancellation probability satisfies a criterion. Attachment 1 describes a churn prediction system.

[0113] [Appendix 3] The information regarding the contract for the communication line is at least one of the usage period of the terminal device used in the contract, the period since the contract renewal, or the period since the contract start. 3. The churn prediction system according to claim 1 or 2.

[0114] [Appendix 4] the acquisition means acquires a result of selecting at least one of the output reasons for prediction, the output means outputs the contract with a probability of cancellation due to the selected predicted reason that is equal to or greater than a standard. Attachment 1 to 3: A churn prediction system.

[0115] [Appendix 5] the prediction means predicts the cancellation probability for each of a plurality of communication lines, including communication lines contracted by the same subscriber, based on the usage status of each of the communication lines; the output means outputs the cancellation probability and the predicted reason in association with the subscriber. Attachment 1 to 3: A churn prediction system.

[0116] [Appendix 6] the output means outputs a summary value based on the churn probability of each communication line for each subscriber. Attachment 5 describes a churn prediction system.

[0117] [Appendix 7] the acquiring means acquires a result of selecting at least one contractor from the plurality of contractors outputted; the output means outputs the cancellation probability of the communication line subscribed to by the selected subscriber and the predicted reason. 6. The churn prediction system described in Appendix 6.

[0118] [Appendix 8] the output means outputs, in a list indicating the churn probability and the value of each item of the prediction reason for each communication line, an item whose influence on the churn probability meets a criterion, in a manner emphasizing it more than other items; Attachment 1 to 7, the churn prediction system.

[0119] [Appendix 9] The output means further outputs a recommended action for the contract for which the cancellation probability satisfies a criterion. Attachment 1 to 8. A churn prediction system according to any one of claims 1 to 8.

[0120] [Appendix 10] the prediction means predicts the churn probability for each group or each communication line that has contracted a plurality of communication lines, and predicts the churn probability using the churn prediction model corresponding to the prediction target; 10. The churn prediction system of any one of appendixes 1 to 9.

[0121] [Appendix 11] further comprising a generation means for generating the churn prediction model by learning the relationship between the usage status and whether or not a communication line has been churn. 11. The churn prediction system of any one of appendixes 1 to 10.

[0122] [Appendix 12] the generation means updates the churn prediction model by re-learning the relationship between the predicted usage status of the communication line and actual data on whether or not there is a churn. 12. The churn prediction system described in Appendix 11.

[0123] [Appendix 13] Acquire data on the usage status of the communication line, including information on the contract for the communication line; predicting the probability of cancellation of the communication line based on the acquired data on the usage status using a cancellation prediction model that has learned the relationship between the usage status of the communication line and whether or not the communication line has been cancelled; outputting the churn probability and a prediction reason for predicting the churn probability; Churn prediction methods.

[0124] [Appendix 14] A process of acquiring data on the usage status of the communication line, including information on the contract for the communication line; a process of predicting the probability of cancellation of the communication line based on the acquired data on the usage status using a cancellation prediction model that has learned the relationship between the usage status of the communication line and whether or not the communication line is canceled; a process of outputting the churn probability and a prediction reason for predicting the churn probability; A program recording medium on which a cancellation prediction program for causing a computer to execute the above is recorded.

[0125] The present invention has been described above using the above-described embodiment as an example. However, the present invention is not limited to the above-described embodiment. In other words, the present invention can be applied in various aspects that can be understood by a person skilled in the art within the scope of the present invention. [Explanation of symbols]

[0126] 10. Cancellation Prediction System 11 Acquisition Department 12 Prediction Department 13 Output section 14 Prediction model generation unit 15 Storage section 20 Terminal equipment 30. Cancellation Prediction System 31 Acquisition Department 32 Output section 40 Terminal Equipment 100 Cancellation Prediction System 101 Acquisition Department 102 Prediction Department 103 Output section 200 computers 201 CPU 202 memory 203 Storage device 204 Input / Output Interface 205 Communication I / F

Claims

1. An acquisition means for acquiring data on the usage status of a plurality of communication lines, including communication lines contracted by the same subscriber, the data including at least the usage period of a terminal device used on each of the plurality of communication lines; a prediction means for using a churn prediction model based on a decision tree that has learned the relationship between the usage status of a communication line and whether or not the communication line has been churn, for each of a plurality of communication lines including communication lines contracted by the same subscriber, using the acquired usage status data as input data, predicting a churn probability based on a predicted value that the communication line corresponding to the terminal device will be classified as churn, and extracting, as a prediction reason, a condition for classification that affects the prediction of the churn probability in the decision tree; an output means for outputting a total value based on the churn probability of each communication line for each said contractor, and when a contractor is selected by an operator, outputting the predicted churn probability and the extracted prediction reason for the communication line contracted by the selected contractor; A cancellation prediction system equipped with

2. The data on the usage status of the communication line further includes at least one of a period from contract renewal and a period from contract initiation. The churn prediction system according to claim 1 .

3. the acquisition means acquires a selection result of at least one of the reasons for prediction input by an operator through an operation of a terminal device to which the reasons for prediction are output, the output means outputs the contract with a probability of cancellation due to the predicted reason indicated by the selection result that is equal to or greater than a standard. The cancellation prediction system according to claim 1 or 2.

4. the acquiring means acquires a selection result of selecting at least one contractor from the plurality of contractors, the selection result being input by an operator's operation on a terminal device to which the aggregated value for each contractor is output; the output means outputs the cancellation probability of the communication line subscribed to by the selected subscriber and the predicted reason. The cancellation prediction system according to any one of claims 1 to 3.

5. the output means outputs a list showing the churn probability and the value of each item of the prediction reason for each communication line, emphasizing items whose influence on the churn probability meets a criterion more than other items; The cancellation prediction system according to any one of claims 1 to 4.

6. A computer comprising: acquiring, for each of a plurality of communication lines including communication lines contracted by the same contractor, data on the usage status of the communication lines, including at least the usage period of a terminal device used on each of the plurality of communication lines; Using a churn prediction model based on a decision tree that has learned the relationship between the usage status of a communication line and whether or not the communication line has been churn, the acquired usage status data is used as input data for each of a plurality of communication lines, including communication lines contracted by the same subscriber, to predict a churn probability based on a predicted value that the communication line corresponding to the terminal device will be classified as churn, and extracting, as prediction reasons, case classification conditions that affect the prediction of the churn probability in the decision tree; outputting a summary value based on the churn probability of each communication line for each said subscriber, and when a subscriber is selected by an operator, outputting the predicted churn probability and the extracted predicted reason for the communication line contracted by the selected subscriber; Churn prediction methods.

7. A process for acquiring data on the usage status of a plurality of communication lines, including communication lines contracted by the same subscriber, the data including at least the usage period of a terminal device used on each of the plurality of communication lines; a process of using a churn prediction model based on a decision tree that has learned the relationship between the usage status of a communication line and whether or not the communication line has been churn, using the acquired usage status data as input data for each of a plurality of communication lines including communication lines contracted by the same subscriber, to predict a churn probability based on a predicted value that the communication line corresponding to the terminal device will be classified as churn, and extracting, as prediction reasons, categorization conditions that affect the prediction of the churn probability in the decision tree; a process of outputting a summary value based on the churn probability of each communication line for each said contractor, and when a contractor is selected by an operator, outputting the predicted churn probability and the extracted predicted reason for the communication line contracted by the selected contractor; A cancellation prediction program that causes a computer to execute the above.

Images