Bankruptcy risk factor identification system and bankruptcy risk factor identification method

Abstract

This invention provides a system and method for identifying bankruptcy risk factors using a machine learning model generated from a large amount of accounting data. [Solution] The bankruptcy risk factor identification system 10 includes an input unit 13 that receives accounting data of a target company provided from an accounting system, a storage unit 12 that stores the accounting data of the target company and accounting data of multiple companies collected in advance, an indicator selection unit that selects financial indicators that are bankruptcy risk factors, a model generation unit 11b that generates a machine learning model with financial indicators as explanatory variables and bankruptcy or non-bankruptcy as the dependent variable, a bankruptcy determination unit that applies the accounting data of the target company to the machine learning model and determines whether the target company is bankrupt or not, and a simulation unit 11c that receives input from the user to change the values ​​of the explanatory variables, applies them to the machine learning model, and causes the bankruptcy determination unit to execute a process to determine whether it is bankrupt or not, and a control unit 11 that outputs the determination result to a display means in a display format in which branching paths can be identified.

Description

Technical Field

[0001] The present invention relates to a bankruptcy risk factor identification system and a bankruptcy risk factor identification method. More specifically, the present invention relates to a bankruptcy risk factor identification system and a bankruptcy risk factor identification method that identify bankruptcy risk factors using a machine learning model generated based on a large amount of accounting data.

Background Art

[0002] The bankruptcy of a company largely depends on the decision-making of the management and the financial situation. Conventionally, the evaluation of bankruptcy risk has mainly been based on the analysis of financial indicators and the rules of thumb of analysts. For this reason, techniques for predicting the bankruptcy probability or the probability of suspension or abolition of business of a specific company using a statistical model have been proposed.

[0003] For example, Patent Document 1 discloses a technique for generating a statistical model based on data of predetermined items selected based on the knowledge of experts from the investigation report information of a large number of companies created by a credit investigation company, and applying the data of the same items of a specific company to this model to calculate the probability that the company will suspend or abolish its business within a predetermined period. Further, Patent Document 2 discloses a technique for constructing a learned model for accurately predicting the occurrence of an event (such as bankruptcy) of a company based on quantitative data (such as financial statement data) and qualitative data (such as various company information) of a plurality of companies.

[0004] However, these conventional techniques focus on risk assessment (calculation of bankruptcy probability) from the perspective of stakeholders such as business partners and financial institutions of a specific company, and do not identify risk factors that affect the bankruptcy of the specific company, which is the party, from the perspective of the management of the specific company, and do not provide useful information for making appropriate decisions for business improvement and bankruptcy avoidance.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

[0006] The present invention has been made in view of the above-mentioned prior art, and the object of the present invention is to provide a bankruptcy risk factor identification system and a bankruptcy risk factor identification method that support management decision-making by identifying the point at which a company will go bankrupt using a trained model that uses specific financial indicators generated by machine learning using a large amount of accounting data collected from a large number of companies as explanatory variables. [Means for solving the problem]

[0007] A bankruptcy risk factor identification system according to one aspect of the present invention, made to achieve the above objective, is an information processing system that identifies bankruptcy risk factors based on accounting data, wherein the computer constituting the information processing system includes: an input unit that receives accounting data of a target company provided from a predetermined accounting system; a storage unit that stores the accounting data of the target company and accounting data of multiple companies collected in advance; an indicator selection unit that uses the accounting data of multiple companies to select financial indicators that become bankruptcy risk factors by statistical methods or deep learning; a model generation unit that generates a machine learning model with the selected financial indicators as explanatory variables and bankruptcy or non-bankruptcy as the dependent variable; and the generated The system is characterized by comprising: a bankruptcy determination unit that applies the accounting data of the target company to a machine learning model to determine whether the target company is bankrupt or not; a simulation unit that receives input from a user via an input unit to change the value of a financial indicator corresponding to any of the explanatory variables of the machine learning model, and causes the bankruptcy determination unit to execute a process to determine whether the target company is bankrupt or not by applying the input value of the financial indicator to the machine learning model; and a control unit that controls the output of the bankruptcy determination result of the bankruptcy determination of the target company performed by the bankruptcy determination unit to a display means in a display format that allows the determination process to be identified, and controls the entire information processing system.

[0008] A method for identifying bankruptcy risk factors according to one aspect of the present invention, made to achieve the above objective, is a method for identifying bankruptcy risk factors based on accounting data using an information processing system, and is characterized by including the steps of: a control unit of a computer constituting the information processing system receiving accounting data of a target company provided from a predetermined accounting system via an input unit; storing the received accounting data of the target company in a storage unit; selecting financial indicators that are bankruptcy risk factors using a statistical method or deep learning with respect to accounting data of multiple companies previously stored in the storage unit; generating a machine learning model with the selected financial indicators as explanatory variables and bankruptcy or non-bankruptcy as the dependent variable; applying the accounting data of the target company to the generated machine learning model to determine whether the target company is bankrupt or not; receiving input from a user to change the value of a financial indicator corresponding to any of the explanatory variables of the machine learning model, and executing a process to apply the input value of the financial indicator to the machine learning model to determine whether the target company is bankrupt or not; and outputting the executed determination result of whether the target company is bankrupt or not to a display means in a display format that allows the determination process to be identified. [Effects of the Invention]

[0009] The bankruptcy risk factor identification system of the present invention can identify financial indicators that constitute bankruptcy risk factors and their thresholds for a target company by utilizing accounting data of the company provided from the accounting system. Furthermore, since the bankruptcy risk factor identification system of the present invention is configured so that the user can change the values ​​input to the explanatory variables of the trained model used, the impact of each individual bankruptcy risk factor can be simulated individually. [Brief explanation of the drawing]

[0010] [Figure 1] This figure shows the overall configuration of a bankruptcy risk factor identification system according to one embodiment of the present invention. [Figure 2] This block diagram shows an example of the functional configuration of a bankruptcy risk factor identification system according to one embodiment of the present invention. [Figure 3] This figure shows an example of the configuration of a database table provided in the accounting data storage unit. [Figure 4] This flowchart shows the procedure for bankruptcy determination processing using a bankruptcy risk factor identification system according to one embodiment of the present invention. [Figure 5] This figure shows an example of a screen for displaying the tree structure of a trained model and accepting numerical input to each node of the tree. [Figure 6] This flowchart shows the procedure for creating a decision tree model using the bankruptcy risk factor identification system of this embodiment. [Modes for carrying out the invention]

[0011] Hereinafter, specific examples of embodiments for carrying out the present invention will be described in detail with reference to the drawings.

[0012] Figure 1 shows the overall configuration of a bankruptcy risk factor identification system according to one embodiment of the present invention.

[0013] As shown in Figure 1, the bankruptcy risk factor identification system 10 according to one embodiment of the present invention is a system composed of a general-purpose computer and includes a control unit 11 that executes a predetermined program and has a central processing unit (CPU), ROM and RAM (not shown), etc., a storage unit 12 consisting of an SSD (Solid State Drive) or HDD (Hard Disk Drive) etc. for storing various programs and data, an input unit 13 that receives commands from users and input of various data, a display unit 14 that displays input / output information and calculation results on an image display means, an output unit 15 that outputs data to a printer that writes to paper media and / or a data writing device that writes to a computer-readable recording medium, etc., and a communication unit 16 that connects to an external client terminal 20 and / or an accounting system 30 etc. via a predetermined communication network. The input unit 13 and the display unit 14 are configured as external input / output devices 17 that each include an input means and an image display means connected to the computer, or are configured to be integrated into the computer.

[0014] In the bankruptcy risk factor identification system 10 according to this embodiment, the control unit 11 executes the OS (Operating System) and a predetermined program on the central processing unit (CPU) to realize a plurality of functional units, which will be described later, and also controls each of these functional units.

[0015] The storage unit 12 is provided with storage areas corresponding to the content to be stored: an accounting data storage unit 12a, a learning data storage unit 12b, an analysis model storage unit 12c, and a judgment data storage unit 12d. The storage unit 12 is not limited to being composed of a single storage device, but may be composed of multiple storage devices connected by a network. That is, the storage unit 12 may be built into a computer that constitutes the bankruptcy risk factor identification system 10, or it may be composed of an external data server connected by a communication network.

[0016] The accounting data storage unit 12a stores accounting data and corporate information of multiple companies. The accounting data and corporate information stored in the accounting data storage unit 12a are not only for the companies (hereinafter referred to as target companies) targeted to identify bankruptcy risk factors, but also accounting data and corporate information of a large number of companies with different industries collected and accumulated by economic and financial data service companies, credit investigation agencies, accounting firms, etc. are pre-accumulated.

[0017] Here, the accounting data can be created, for example, using general accounting software installed in a computer constituting the bankruptcy risk factor identification system 10 or in an external client terminal 20, or provided from an external accounting system 30. The accounting data created by the above-mentioned accounting software or provided from outside is stored in the accounting data storage unit 12a in the form of, for example, CSV data format.

[0018] In addition, when the accounting data is created in an external client terminal 20 or provided from an external accounting system 30, the accounting data can be provided (transmitted) directly from the external client terminal 20 or the external accounting system 30 to the bankruptcy risk factor identification system 10 via a network connection, or provided (input) to the bankruptcy risk factor identification system 10 in a form stored on a computer-readable storage medium.

[0019] The learning data storage unit 12b stores data (referred to as learning data) selected for use in training the machine learning model from the accounting data and corporate information of all companies stored in the accounting data storage unit 12a. The learning data can be selected, for example, so that bankrupt companies and non-bankrupt companies are included at a predetermined ratio during the same economic environment period by industry, or created by applying a predetermined unbalanced data countermeasure method. The data items included in the learning data can be specified or changed by the administrator or user of this system 10. Note that the learning data may include training data and test data.

[0020] The analysis model storage unit 12c stores a machine learning model (hereinafter referred to as the "trained model") that has been trained using the training data described above to determine (predict) whether a company will go bankrupt or not. In this embodiment, an example in which a decision tree model is used as the machine learning model will be described. However, the machine learning model used is not limited to this. For example, Random Forest, XGBoost, etc. may be used.

[0021] The judgment data storage unit 12d stores the results obtained by applying the accounting data of the target company (i.e., the target company) to the trained model for identifying bankruptcy risk factors, and the results obtained when some of the numerical values ​​in the target company's accounting data applied to the trained model are changed.

[0022] The bankruptcy risk factor identification system 10 according to the embodiment of the present invention described above is not limited to a configuration consisting of a single computer, but may also consist of multiple computers connected to a network. Alternatively, the bankruptcy risk factor identification system 10 may be configured in a cloud-based manner, which responds to requests from external client terminals 20, performs bankruptcy risk factor identification processing, and returns the processing results to the client terminals.

[0023] Next, the functional configuration of the bankruptcy risk factor identification system 10 according to one embodiment of the present invention will be described.

[0024] Figure 2 is a block diagram showing an example of the functional configuration of a bankruptcy risk factor identification system according to one embodiment of the present invention.

[0025] As shown in Figure 2, the bankruptcy risk factor identification system 10 according to one embodiment of the present invention includes a data management unit 11a, a model creation unit 11b, and a simulation unit 11c as functional units realized by having the CPU of the control unit 11 execute a predetermined program (various programs for identifying bankruptcy risk factors). Furthermore, the model creation unit 11b includes an indicator selection unit, and the simulation unit 11c includes a bankruptcy determination unit.

[0026] The data management unit 11a organizes and manages accounting data and corporate information of numerous companies stored in the storage unit 12, and also performs the process of creating training data.

[0027] Specifically, the data management unit 11a inputs the large amount of corporate accounting data and corporate information (collected raw data) stored in the accounting data storage unit 12a of the storage unit 12 into the corresponding data item cells (fields) of a data table (called a database table) pre-established in the accounting data storage unit 12a, one for each company. It then assigns an identifying corporate ID to each data row (record) consisting of these cells, and registers (saves) them in association with class information indicating whether the company is bankrupt or not (surviving). Finally, it sorts the rows so that each company's data row is grouped by industry on the data table.

[0028] Figure 3 shows an example of the configuration of a database table provided in the accounting data storage unit 12a.

[0029] The database table 300 shown in Figure 3 is organized by industry and, associated with the company ID (identification number) 310, includes the survey year (year of bankruptcy) 320, a class (CLASS) 330 indicating bankruptcy (B) or non-bankruptcy (N) which will be the target variable of the trained model, and multiple financial indicators or company information (X1, X2, ..., X) which may be explanatory variables of the model. n The system has a configuration in which data rows (records) are arranged by combining data sets 340 containing ). The types and number of data items included in data set 340 can be specified by the administrator or user of this system 10. The example configuration of the database table 300 in Figure 3 includes data items X1, X2, ..., X 13 While examples are shown where predetermined financial indicators (financial ratios) are set, the present invention is not limited thereto.

[0030] Even when accounting data and company information of a company that is subject to bankruptcy risk factor identification are entered into the data management unit 11a, it registers the data of the specified data items in the database table 300 as described above.

[0031] The data management unit 11a further creates a dataset (training data) from the database table 300 for use in training a machine learning model, according to instructions input from an administrator or user, and stores it in the training data storage unit 12b.

[0032] The model creation unit 11b generates a trained machine learning model (pre-trained model) by applying the training data stored in the training data storage unit 12b to a pre-configured machine learning model. In this embodiment, the program for generating the machine learning model can, for example, use the rpart package of the "R language" or the scikit-learn library of "Python" as statistical analysis software. However, the statistical analysis programs applicable in this invention are not limited to these. Furthermore, since past research has shown that the causes of bankruptcy tend to differ depending on the industry, the generated pre-trained models are generated by industry in this embodiment, but this invention is not limited to this.

[0033] The simulation unit 11c uses a pre-trained model corresponding to the industry of the company (target company) for which bankruptcy risk factors are identified to determine whether or not the company will go bankrupt. Specifically, the bankruptcy determination unit of the simulation unit 11c reads each data of the target company corresponding to the explanatory variables (also called input variables) of the pre-trained model from the database table 300 of the accounting data storage unit 12a, applies (inputs) this data to the pre-trained model, and performs the process of determining whether or not the target company will go bankrupt.

[0034] Furthermore, the simulation unit 11c receives instructions from the user via the input unit 13 to specify a desired explanatory variable from among the explanatory variables of a trained model and to apply (input) data modified from the original data to that explanatory variable. The simulation unit then instructs the bankruptcy determination unit to apply the received data (modified data) to the model and perform a determination process to determine whether or not the company will go bankrupt, thereby simulating whether or not the determination result of bankruptcy or non-bankruptcy changes as a result of the modification.

[0035] Next, the bankruptcy determination process using the bankruptcy risk factor identification system according to one embodiment of the present invention will be explained with reference to Figures 2 and 4.

[0036] Figure 4 is a flowchart showing the procedure for bankruptcy determination processing using a bankruptcy risk factor identification system according to one embodiment of the present invention.

[0037] Referring to Figure 4, after the bankruptcy risk factor identification system 10 is activated, the bankruptcy risk factor identification system 10 receives accounting data and company information of the target company via the input unit 13 (S100). The accounting data includes, for example, financial indicator data for the past year of the target company and journal entry data created using other common accounting software, and each data is provided in CSV format. The company information includes, for example, the target company's company code, industry, trade name, representative, address, listing status, date of establishment, capital, number of employees, etc.

[0038] When accounting data for a target company is created and provided (output) using accounting software implemented in a computer constituting the bankruptcy risk factor identification system 10, the user inputs a command via the input unit 13 to specify the storage unit 12 as the output destination for the accounting data created by the accounting software. Once the output destination is specified as the storage unit 12, the control unit 11 transmits the accounting data output from the accounting software along with company information to the data management unit 11a, and causes the data management unit 11a to execute a process to register (save) the data in a predetermined data table (database table 300) constructed in the accounting data storage unit 12a of the storage unit 12 (step S110).

[0039] On the other hand, if the accounting data of the target company is provided from a client terminal 20 managed by the user via a predetermined communication network, the user sends a command from the client terminal 20 to the bankruptcy risk factor identification system 10 requesting the registration of accounting data and company information. The control unit 11 of the bankruptcy risk factor identification system 10 receives the accounting data and company information via the communication unit 16 and causes the data management unit 11a to execute step S110, which registers (saves) the data in a predetermined data table in the accounting data storage unit 12a of the storage unit 12.

[0040] In step S110, which registers the received accounting data and company information into a predetermined data table in the accounting data storage unit 12a, the data management unit 11a distributes the received accounting data and company information to the data items included in the predetermined data table (database table 300) of the accounting data storage unit 12a and registers (saves) them. At this time, if the received accounting data does not contain data corresponding to a specific financial indicator included in the database table 300, the data management unit 11a may perform a process to calculate and register the numerical value of the financial indicator that is not included in the received accounting data using the data of the relevant account included in the received accounting data.

[0041] Furthermore, users can provide accounting data and company information in a form stored on a computer-readable recording medium and input it via the input unit 13.

[0042] Subsequently, the control unit 11 of the bankruptcy risk factor identification system 10 selects a pre-trained model suitable for the target company's industry from among multiple pre-trained models stored in the analysis model storage unit 12c of the storage unit 12, based on the company information of the target company received in step S100 (step S120).

[0043] Next, the control unit 11 instructs the simulation unit 11c to acquire (read) the selected trained model from the analysis model storage unit 12c, and the simulation unit 11c causes the bankruptcy determination unit to execute a process to determine whether or not the target company will go bankrupt using the acquired trained model (step S130). At this time, as a default operation, the bankruptcy determination unit reads the accounting data or company information of the target company that corresponds to the explanatory variables of the trained model from the database table 300 of the accounting data storage unit 12a and applies (inputs) it to the trained model.

[0044] In this embodiment, before executing the bankruptcy determination process in step S130, the control unit 11 displays an image of the tree structure of the selected trained model on the display unit 14, allowing the user to visually confirm the node configuration of the model. The control unit 11 then accepts input from the user via the input unit 13 to confirm or change the data to be applied (input) to each node (explanatory variable) included in the trained model (step S125).

[0045] Figure 5 shows an example of a screen for receiving data input to each node (explanatory variable) of the trained model displayed on the display unit 14 in step S125. The tree shown in Figure 5 is an example with one intermediate layer, but this is for illustrative purposes only. The bankruptcy risk factor identification system according to the present invention is configured so that the number of layers (tree depth) can be set or changed by the system administrator or user.

[0046] The display screen 500 shown in Figure 5 displays input fields (501-503) for each node (x1, x2, x3) of the decision tree corresponding to the selected pre-trained model, allowing the user to specify the data (values) they want to try (simulate) instead of the actual data (real data) of the target company. After the user enters the desired data (value) into the input field of the node they want to change, they press the "Execute" button 520 displayed on the display screen 500, and the entered data (value) is reflected in the pre-trained model. For example, if the return on total capital is set to node x1 of the selected pre-trained model, and the actual data of the target company is entered into node x1 by default (the actual data is displayed in input field 501), the user can check whether the determination of bankruptcy or not changes by entering (overwriting) the desired value of the return on total capital into input field 501 of node x1 and pressing the "Execute" button 520 to check the impact of changes in the return on total capital.

[0047] Additionally, the display screen 500 includes a field 510 where the company ID is displayed. After the user confirms the company ID, if there is no need to change the data entered in each node (explanatory variable), they can press the execute button 520 without entering anything in the input fields (501-503) of each node (x1, x2, x3) (actual data is displayed by default). The actual data of the target company corresponding to the company ID will then be applied (input) to the trained model in each node, and the classification process will be performed.

[0048] In other words, if there is no data input from the user in step S125 (default setting), the simulation unit 11c compares the branching condition (threshold) z1 for the financial indicator set for the first explanatory variable x1 (root node) of the selected trained model with the corresponding numerical data of the financial indicator included in the accounting data of the target company to determine the branching node. For example, if the branching destination is node x2, the simulation unit 11c compares the branching condition (threshold) z2 for the financial indicator of the second explanatory variable set for node x2 with the corresponding numerical data of the financial indicator included in the accounting data of the target company to determine the next branching node, and repeats this step to determine whether the target company is classified as bankrupt or not at the final node.

[0049] The control unit 11 transmits the determination result of whether the target company is bankrupt or not, made by the simulation unit 11c, to the output unit 15, and displays the determination process (branching path) on the display unit 14 in a visually identifiable display format (step S140). For example, in the display screen 500 of Figure 5, the determination process of the target company may be shown so that the path from the root node x1 to the final classified node (leaf node) y1 and the block of leaf node y1 are highlighted. The control unit 11 may also display branching conditions (thresholds) at each node on the display screen 500.

[0050] Subsequently, the control unit 11 receives input to specify whether to terminate the process (Y) or to execute (re-determine) under different conditions (N) (step S150). If an instruction to re-determine under different conditions is received in step S150, the control unit 11 returns to step S125 and receives input from the user to specify data (values) to be substituted into one of the explanatory variables (decision tree nodes) included in the trained model displayed on the display screen 500 in Figure 5, and then substitutes the received data (values) into the trained model to execute the process of determining whether the target company is bankrupt or not (step S130).

[0051] Therefore, in addition to being able to determine whether a company is bankrupt or not based on its current accounting data (financial indicators), users can simulate how the determination will change depending on how the values ​​of the explanatory variables (financial indicators) included in the trained model change. This allows users to identify the threshold values ​​at which specified financial indicators become risk factors leading to bankruptcy.

[0052] Furthermore, users can determine whether, in the case of the target company, the actual values ​​corresponding to any of the explanatory variables (financial indicators) included in the trained model (i.e., financial indicators highly correlated with bankruptcy) on the bankruptcy branching path displayed on the screen are dangerous in relation to the threshold of the branching point, given the current state of the business.

[0053] In the bankruptcy risk factor identification method according to this embodiment, the bankruptcy risk factor identification system 10 executes various programs for identifying bankruptcy risk factors and automatically proceeds through a series of steps from step S100 to step S150. Therefore, the user's operation is simplified to two commands: an output command for accounting data and company information to the accounting software or accounting system described above, and an input command indicating whether or not to change the input data for the explanatory variables of the pre-trained model selected by the bankruptcy risk factor identification system 10. Thus, the user can obtain a judgment result based on the pre-trained model without requiring specialized knowledge. Furthermore, the bankruptcy risk factor identification system 10 according to this embodiment can also generate a machine learning model by allowing the user to specify a particular financial indicator from among the financial indicators included in the received accounting data. In this case, the system can further include a configuration that accepts user input to set the financial indicator to be used for discrimination.

[0054] Next, we will describe a method for generating a machine learning model used in a bankruptcy risk factor identification system and method according to one embodiment of the present invention, using a large amount of accounting data and company information.

[0055] In this embodiment, when a decision tree model is adopted as the machine learning model, multiple financial indicators included in a large amount of pre-collected corporate accounting data are used to select financial indicators that have a high correlation with bankruptcy from among the multiple financial indicators and place them in the nodes of the decision tree.

[0056] Figure 6 is a flowchart showing the procedure for creating a decision tree model using the bankruptcy risk factor identification system of this embodiment.

[0057] Referring to Figure 6, the control unit 11 of the bankruptcy risk factor identification system 10 according to one embodiment of the present invention, for example, when it receives a command from an administrator to create a machine learning model, activates the model creation unit 11b and has it read the data of the class indicating whether each company in each industry is bankrupt or not, and each financial indicator, stored in the database table 300 of the accounting data storage unit 12a (step S200).

[0058] The indicator selection unit of the model creation unit 11b uses the imported data on the bankruptcy or non-bankruptcy class and each financial indicator for each company in each industry to select financial indicators that are highly correlated with bankruptcy using known statistical methods or deep learning (step S210). The model creation unit 11b may also display the selected financial indicators in order of their correlation with bankruptcy on the display unit 14 in a predetermined display format, for example, a list format (not shown).

[0059] Subsequently, the model creation unit 11b executes a statistical analysis program to automatically create a decision tree model for each industry, with a class indicating whether a company is bankrupt or not as the target variable, according to the branching depth of the decision tree, the minimum number of samples per node, and the number of explanatory variables set in advance by the administrator or user (step S220). It then reads the training data for the relevant industry from the training data storage unit 12b and uses the training data for that industry to determine the financial indicators and thresholds to be set for each node of the decision tree (step S230).

[0060] In step S220 shown in Figure 6, the model creation unit 11b generates decision tree models for each industry because past research has shown that different financial indicators have a greater impact on bankruptcy depending on the industry. This is done to avoid unnecessarily expanding the data range and to improve the efficiency of model derivation and the accuracy of the models.

[0061] In this embodiment, the statistical analysis program used to automatically create the decision tree model is not particularly limited, but for example, the rpart package in the "R language" or the scikit-learn library in "Python" can be used. The financial indicators set in the root node may be financial indicators that are generally considered to have a high correlation with corporate bankruptcy by credit rating agencies or accounting firms, or they may be determined from the financial indicators selected in step S210 based on the calculation results of Gini impurity or entropy.

[0062] Once the financial indicators and thresholds to be set for each node of the decision tree have been determined, the decision tree is complete, and it is automatically confirmed that the model accuracy has reached a predetermined standard value, the model creation unit 11b saves the completed industry-specific decision tree model to the analysis model storage unit 12c (step S240). Similarly, industry-specific decision tree models are created for other industries and saved to the analysis model storage unit 12c using the same procedure.

[0063] As described above, the bankruptcy risk factor identification system 10 according to one embodiment of the present invention automatically generates a trained model for determining whether a company will go bankrupt or not from a large amount of company accounting data and company information, applies the accounting data of the target company to the generated trained model to determine whether the company will go bankrupt or not, and in addition, the user can change the explanatory variables of the trained model, i.e., the numerical values ​​set for factors that are highly correlated with bankruptcy, and try to make a determination (simulation), so that the user can identify financial indicators that are bankruptcy risk factors and their thresholds (boundary values).

[0064] Therefore, the bankruptcy risk factor identification system according to one embodiment of the present invention can simulate the avoidance of bankruptcy risk factors by changing the financial indicators of a target company, and can provide guidance for managers when making decisions.

[0065] The above-mentioned method for identifying bankruptcy risk factors can be implemented using computer-executable programs written in various programming languages, and can be stored and distributed on computer-readable recording media such as ROM, EEPROM, EPROM, flash memory, CD-ROM, CD-RW, DVD, SD card, and USB memory.

[0066] Although embodiments of the present invention have been described in detail above with reference to the drawings, the present invention is not limited to the embodiments described above, and can be modified and implemented in various ways without departing from the technical scope of the present invention. [Explanation of Symbols]

[0067] 10. Bankruptcy Risk Factor Identification System 11 Control Unit 11a Data Management Department 11b Model Creation Section 11c Simulation Department 12 Storage section 12a Accounting data storage unit 12b Learning data storage unit 12c Analysis Model Memory Unit 12d Judgment Data Storage Unit 13 Input section 14 Display section 15 Output section 16 Communications Department 17 External Input / Output Devices 20 client terminals 30 External Accounting Systems

Claims

1. An information processing system that identifies bankruptcy risk factors based on accounting data, The computer constituting the aforementioned information processing system is An input unit that receives accounting data of the target company provided from a designated accounting system, A storage unit that stores the accounting data of the aforementioned target company and the accounting data of multiple companies that have been collected in advance, An indicator selection unit that uses accounting data from the aforementioned multiple companies to select financial indicators that are risk factors for bankruptcy using statistical methods or deep learning, A model generation unit generates a machine learning model in which the selected financial indicators are explanatory variables and bankruptcy or non-bankruptcy is the dependent variable. A bankruptcy determination unit that applies the accounting data of the target company to the generated machine learning model to determine whether the target company is bankrupt or not, A simulation unit that receives input from a user via the input unit to change the value of a financial indicator corresponding to one of the explanatory variables of the machine learning model, and causes the bankruptcy determination unit to execute a process to determine whether the target company is bankrupt or not by applying the input value of the financial indicator to the machine learning model, A bankruptcy risk factor identification system characterized by comprising: a control unit that controls the output of the bankruptcy or non-bankruptcy determination result of the target company, performed by the bankruptcy determination unit, to a display means in a display format that allows for the identification of branching paths, and a control unit that controls the entire information processing system.

2. The aforementioned machine learning model is a decision tree model, The bankruptcy risk factor identification system according to claim 1, characterized in that the indicator selection unit selects financial indicators that constitute bankruptcy risk factors based on the feature importance at each node of the decision tree model calculated using the accounting data of the multiple companies.

3. The bankruptcy risk factor identification system according to claim 2, characterized in that the financial indicators set for each node of the decision tree model are selected by deep learning using the accounting data of the multiple companies.

4. A method for identifying bankruptcy risk factors based on accounting data using an information processing system, The control unit of the computer constituting the aforementioned information processing system, The stage of receiving accounting data of the target company provided from a designated accounting system via an input unit, The step of storing the accounting data of the target company received in the storage unit, The steps include selecting financial indicators that are risk factors for bankruptcy using statistical methods or deep learning, based on accounting data of multiple companies pre-stored in the aforementioned memory unit, The steps include: generating a machine learning model with the selected financial indicators as explanatory variables and bankruptcy or non-bankruptcy as the dependent variable; The steps include applying the accounting data of the target company to the generated machine learning model to determine whether the target company is bankrupt or not, The process involves receiving input from a user to change the value of a financial indicator corresponding to one of the explanatory variables of the machine learning model, applying the input value of the financial indicator to the machine learning model, and executing a process to determine whether the target company is bankrupt or not. A method for identifying bankruptcy risk factors, characterized by including the step of outputting the result of the determination of whether the target company is bankrupt or not, which has been performed, to a display means in a display format that allows for the identification of branching paths.

5. The aforementioned machine learning model is a decision tree model, The method for identifying bankruptcy risk factors according to claim 4, characterized in that the step of selecting financial indicators that constitute bankruptcy risk factors includes a step in which the control unit selects financial indicators that constitute bankruptcy risk factors based on the feature importance at each node of the decision tree model calculated using the accounting data of the multiple companies.

6. The method for identifying bankruptcy risk factors according to claim 5, further comprising the step of selecting financial indicators to be set for each node of the decision tree model by deep learning using accounting data of the multiple companies.

Images