Method and device for determining default score based on logistic regression model, and program product

Abstract

The embodiment of the invention provides a method and a device for determining a default score based on a logistic regression model, and a program product. The method comprises the following steps: determining target credit data of a target object in a first time period; constructing a target feature variable table of the target credit data; and inputting the target feature variable table into the target logistic regression model to obtain a default score of the target object output by the target logistic regression model in the second time period. Through the method and the device, the problem of relatively low evaluation accuracy of the default score of the credit product applied by the common enterprise in the related technology is solved, and the effect of improving the evaluation accuracy of the default score of the credit product applied by the enterprise is achieved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular, to a method and device for determining a default score based on a logistic regression model, and a program product. Background Art

[0002] At present, the assessment of pre-loan risk of credit business mainly adopts offline manual review or corporate credit rating model for assessment. Manual review requires the account manager to conduct on-site investigation of the business conditions of the enterprise and the credit, financial and other information of the actual controller of the enterprise, and make assessments based on human experience. The data analysis source is single, not timely, and has large errors. The corporate credit rating model conducts risk assessment on the enterprise through data information such as the company's financial statements and financial assets within the bank. However, for enterprises without financial statements (for example, small and micro enterprises), it is impossible to use the corporate credit rating model for default assessment. Summary of the invention

[0003] The embodiments of the present application provide a method, device, and program product for determining a default score based on a logistic regression model, so as to at least solve the problem of low assessment accuracy of default scores for inclusive enterprises applying for credit products in the related art.

[0004] According to one embodiment of the present application, a method for determining a default score based on a logistic regression model is provided, comprising: determining target credit data of a target object within a first time period, wherein the target object is an inclusive enterprise whose credit line is greater than a preset credit line, and the target credit data includes data generated by credit operations performed by the target object within the first time period; constructing a target feature variable table for the target credit data, wherein the target feature variable table includes a plurality of target feature variables; inputting the target feature variable table into a target logistic regression model to obtain a default score of the target object within a second time period output by the target logistic regression model, wherein the second time period is a time period after the first time period.

[0005] In an exemplary embodiment, determining the target credit data of the target object within a first time period includes: obtaining credit data of multiple dimensions generated by the credit operation performed on the target object within the first time period, wherein the credit data of the multiple dimensions include: financial data, historical credit data, loan data, and macroeconomic data within the first time period; performing preprocessing operations on the credit data of the multiple dimensions to obtain the target credit data, wherein the preprocessing operations include: removing duplicate data in the credit data of each dimension according to credit logic business rules, modifying erroneous data in the credit data of each dimension, and filling missing data in the credit data of each dimension, and the credit logic business rules are determined based on historical credit data.

[0006] In an exemplary embodiment, constructing a target characteristic variable table of the target credit data includes: performing numerical processing on the target credit data to obtain multiple characteristic variables of the target credit data; combining the multiple characteristic variables according to the correlation between the multiple characteristic variables to derive multiple combined characteristic variables; performing discretization and normalization operations on the multiple characteristic variables to derive multiple transformed characteristic variables; performing crossover operations on the multiple characteristic variables using leaf nodes in a decision tree model to derive multiple crossed characteristic variables; screening the multiple target characteristic variables from the multiple combined characteristic variables, the multiple transformed characteristic variables and the multiple crossed characteristic variables according to screening indicators, wherein the screening indicators include variable missing rate, significance, correlation, group stability index and information value; integrating the multiple target characteristic variables to obtain the target characteristic variable table.

[0007] In an exemplary embodiment, before the target feature variable table is input into the target logistic regression model to obtain the default score of the target object output by the target logistic regression model, the method includes: selecting target sample feature variables from the sample feature variable table according to the marginal contribution of each sample feature variable to obtain the target sample feature variable table; inputting each of the target sample feature variables in the target sample feature variable table into the original logistic regression model one by one, and detecting the output results of the original logistic regression model one by one; when the output results corresponding to each of the target sample feature variables in the target sample feature variable table meet the preset output results, determining that the target logistic regression model is obtained through training.

[0008] In an exemplary embodiment, when the output results corresponding to each of the target sample characteristic variables in the target sample characteristic variable table meet the preset output results, before determining that the target logistic regression model is trained, the method includes: detecting the sign of the target sample characteristic variable in the target sample characteristic variable table according to a first test rule, wherein the first test rule is determined based on historical default data; calculating the significance score and variance inflation coefficient of the target sample characteristic variable in the target sample characteristic variable table; deleting or replacing the target sample characteristic variables with wrong signs in the target sample characteristic variable table; deleting or replacing the target sample characteristic variables in the target sample characteristic variable table whose significance score is greater than a first threshold; deleting or replacing the target sample characteristic variables in the target sample characteristic variable table whose variance inflation coefficient is greater than a second threshold.

[0009] In an exemplary embodiment, when the output results corresponding to each of the target sample feature variables in the target sample feature variable table meet the preset output results, after determining that the target logistic regression model is obtained through training, the method includes: determining a rolling rate of historical default objects, wherein the rolling rate is used to represent the loan change trend of the historical default objects in the historical time period; determining the first default probability of the historical default objects in the historical time period according to the rolling rate; and determining the target default probability of the historical default objects in a third time period based on the first default probability and the output results, wherein the third time period is a time period after the historical time period.

[0010] In an exemplary embodiment, determining the target default probability of the historical default object in the third time period based on the first default probability and the output result includes: obtaining a preset score threshold matching the first default probability, wherein the first default probability is inversely proportional to the preset score threshold; comparing the default score with the preset score threshold to obtain a comparison result; determining the target default probability based on the comparison result, wherein, when the comparison result shows that the default score is greater than the preset score threshold, the target default probability is a probability less than the first default probability, and when the comparison result shows that the default score is less than the preset score threshold, the target default probability is a probability greater than the first default probability.

[0011] According to one embodiment of the present application, a device for determining a default score based on a logistic regression model is provided, comprising: a first determination module, used to determine target credit data of a target object within a first time period, wherein the target object is an inclusive enterprise whose credit line is greater than a preset credit line, and the target credit data includes data generated by credit operations performed by the target object within the first time period; a first construction module, used to construct a target feature variable table for the target credit data, wherein the target feature variable table includes multiple target feature variables; a first input module, used to input the target feature variable table into a target logistic regression model to obtain a default score of the target object within a second time period output by the target logistic regression model, wherein the second time period is a time period after the first time period.

[0012] In an exemplary embodiment, the above-mentioned first determination module includes: a first acquisition sub-module, used to obtain credit data of multiple dimensions generated by the above-mentioned target object performing credit operations within the above-mentioned first time period, wherein the above-mentioned credit data of multiple dimensions include: financial data, historical credit data, loan data, and macroeconomic data within the above-mentioned first time period; a first execution sub-module, used to perform preprocessing operations on the above-mentioned credit data of multiple dimensions to obtain the above-mentioned target credit data, wherein the above-mentioned preprocessing operations include: removing duplicate data in the credit data of each dimension according to credit logic business rules, modifying erroneous data in the credit data of each dimension, and filling missing data in the credit data of each dimension, and the above-mentioned credit logic business rules are determined based on historical credit data.

[0013] In an exemplary embodiment, the above-mentioned first construction module includes: a first processing submodule, which is used to perform numerical processing on the above-mentioned target credit data to obtain multiple characteristic variables of the above-mentioned target credit data; a first derivation submodule, which is used to combine the multiple characteristic variables according to the correlation between the multiple characteristic variables to derive multiple combined characteristic variables; a second derivation submodule, which is used to perform discretization operations and normalization operations on the multiple characteristic variables to derive multiple transformed characteristic variables; a third derivation submodule, which is used to perform crossover operations on the multiple characteristic variables using leaf nodes in a decision tree model to derive multiple crossed characteristic variables; a first screening submodule, which is used to screen the multiple target characteristic variables from the multiple combined characteristic variables, the multiple transformed characteristic variables and the multiple crossed characteristic variables according to screening indicators, wherein the screening indicators include variable missing rate, significance, correlation, group stability index and information value; a first integration submodule, which is used to integrate the multiple target characteristic variables to obtain the target characteristic variable table.

[0014] In an exemplary embodiment, the above-mentioned device includes: a first selection module, which is used to input the above-mentioned target feature variable table into the target logistic regression model, and before obtaining the default score of the above-mentioned target object output by the above-mentioned target logistic regression model, select the target sample feature variable from the sample feature variable table according to the marginal contribution of each sample feature variable to obtain the target sample feature variable table; a first detection module, which is used to input each of the above-mentioned target sample feature variables in the above-mentioned target sample feature variable table into the original logistic regression model one by one, and detect the output results of the above-mentioned original logistic regression model one by one; a second determination module, which is used to determine that the above-mentioned target logistic regression model is obtained by training when the output results corresponding to each of the above-mentioned target sample feature variables in the above-mentioned target sample feature variable table meet the preset output results.

[0015] In an exemplary embodiment, the above-mentioned device also includes: a first verification module, which is used to detect the signs of the target sample characteristic variables in the above-mentioned target sample characteristic variable table according to the first verification rule before determining that the output results corresponding to each of the above-mentioned target sample characteristic variables in the above-mentioned target sample characteristic variable table meet the preset output results, wherein the above-mentioned first verification rule is determined based on historical default data; a first calculation module, which is used to calculate the significance score and variance inflation coefficient of the target sample characteristic variables in the above-mentioned target sample characteristic variable table; a first processing module, which is used to delete or replace the target sample characteristic variables with wrong signs in the above-mentioned target sample characteristic variable table; a second processing module, which is used to delete or replace the target sample characteristic variables in the above-mentioned target sample characteristic variable table whose significance scores are greater than the first threshold; and a third processing module, which is used to delete or replace the target sample characteristic variables in the above-mentioned target sample characteristic variable table whose variance inflation coefficients are greater than the second threshold.

[0016] In an exemplary embodiment, the above-mentioned device includes: a third determination module, which is used to determine the rolling rate of the historical default object after determining that the above-mentioned target logistic regression model is obtained through training when the output results corresponding to each of the above-mentioned target sample characteristic variables in the above-mentioned target sample characteristic variable table meet the preset output results, wherein the above-mentioned rolling rate is used to represent the loan change trend of the above-mentioned historical default object in the historical time period; a fourth determination module, which is used to determine the first default probability of the above-mentioned historical default object in the above-mentioned historical time period according to the above-mentioned rolling rate; and a fifth determination module, which is used to determine the target default probability of the above-mentioned historical default object in a third time period based on the above-mentioned first default probability and the above-mentioned output results, wherein the above-mentioned third time period is a time period after the above-mentioned historical time period.

[0017] In an exemplary embodiment, the fifth determination module includes: a second acquisition submodule, used to acquire a preset score threshold matching the first default probability, wherein the first default probability is inversely proportional to the preset score threshold; a first comparison submodule, used to compare the default score with the preset score threshold to obtain a comparison result; and a first determination submodule, used to determine the target default probability based on the comparison result, wherein, when the comparison result shows that the default score is greater than the preset score threshold, the target default probability is a probability less than the first default probability, and when the comparison result shows that the default score is less than the preset score threshold, the target default probability is a probability greater than the first default probability.

[0018] According to another embodiment of the present application, a computer program product is provided, including a computer program, wherein the computer program is configured for a processor to execute the steps of any of the above method embodiments.

[0019] According to another embodiment of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to enable a processor to execute the steps in any of the above method embodiments.

[0020] According to another embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor is configured to execute the computer program to perform the steps in any one of the method embodiments.

[0021] Through this application, the target sample characteristic variable table is determined according to the marginal contribution of each sample characteristic variable; when the output results corresponding to each target sample characteristic variable in the target sample characteristic variable table meet the preset output results, the target logistic regression model is obtained. The target credit data of the target object in the first time period is processed using the target logistic regression model to obtain the default score of the target object in the second time period output by the target logistic regression model. Therefore, the problem of low assessment accuracy of default scores for inclusive enterprises applying for credit products in the related art is solved, and the effect of improving the accuracy of default scores for credit products applied by enterprises is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a hardware structure block diagram of a mobile terminal of a method for determining a default score based on a logistic regression model in an embodiment of the present application;

[0023] Figure 2 is a flow chart of a method for determining a default score based on a logistic regression model according to an embodiment of the present application;

[0024] Figure 3 is a flowchart of creating a target logistic regression model in a specific embodiment of the present application;

[0025] Figure 4 It is a structural block diagram of a device for determining a default score based on a logistic regression model according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] The embodiments of the present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0028] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 1 is a hardware structure block diagram of a mobile terminal of a method for determining a default score based on a logistic regression model in an embodiment of the present application. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0029] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as a computer program corresponding to a method for determining a default score based on a logistic regression model in an embodiment of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the mobile terminal via a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0030] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0031] In this embodiment, a method for determining a default score based on a logistic regression model is provided. Figure 2is a flow chart of a method for determining a default score based on a logistic regression model according to an embodiment of the present application. Figure 2 As shown, the process includes the following steps:

[0032] Step S202, determining target credit data of the target object in the first time period, wherein the target credit data includes data generated by the credit operation performed by the target object in the first time period;

[0033] Step S204, constructing a target characteristic variable table of target credit data, wherein the target characteristic variable table includes a plurality of target characteristic variables;

[0034] Step S206, inputting the target characteristic variable table into the target logistic regression model to obtain the default score of the target object output by the target logistic regression model, wherein the target logistic regression model is obtained by training through the following steps: selecting the target sample characteristic variable from the sample characteristic variable table according to the marginal contribution of each sample characteristic variable to obtain the target sample characteristic variable table; inputting each target sample characteristic variable in the target sample characteristic variable table into the original logistic regression model one by one, and detecting the output results of the original logistic regression model one by one; when the output results corresponding to each target sample characteristic variable in the target sample characteristic variable table meet the preset output results, determining that the target logistic regression model is obtained by training;

[0035] Step S208, calculating the migration risk impact multiplier according to the unexpected loss;

[0036] Step S210, determining a target default probability of the target object in a second time period based on the default score, wherein the second time period is a time period after the first time period.

[0037] In this embodiment, the execution subject of the above steps can be a terminal, a server, a specific processor set in the terminal or server, or a processor or processing device set relatively independently from the terminal or server, but is not limited thereto. For example, the existing risk management system of a commercial bank.

[0038] Optionally, the target objects include but are not limited to: micro-enterprises, small enterprises, small micro-enterprises, start-ups, and private enterprises.

[0039] Optionally, the target credit data includes, but is not limited to: the target object's credit history data, the target object's financial status data, the target object's repayment ability data, the target object's guarantee status data, and the target object's debt status data.

[0040] Optionally, the target logistic regression model is used to estimate the probability of occurrence of binary events. For example, using historical performance variables, through regression coefficient values ​​or weights, the possibility of default of customers / accounts within a given time period is predicted. In the regression process of obtaining the target logistic regression model, this embodiment uses the stepwise regression method to combine forward selection and backward selection, gradually put in the best sample feature variables, and remove the worst sample feature variables to achieve the local optimum of the target sample feature variable results. The structure of the target logistic regression model is as follows: constructing linear combinations, constructing logistic functions, probability prediction, and classification decisions. The parameters (weights and bias terms) of the target logistic regression model are usually solved by the maximum likelihood estimation method. In the process of training to obtain the target logistic regression model, gradient descent or other optimization algorithms can be used to minimize the loss function (such as cross entropy loss) to find the optimal parameter values.

[0041] Through the above steps, the target sample characteristic variable table is determined according to the marginal contribution of each sample characteristic variable; when the output results corresponding to each target sample characteristic variable in the target sample characteristic variable table meet the preset output results, the target logistic regression model is obtained. The target credit data of the target object in the first time period is processed using the target logistic regression model to obtain the default score of the target object in the second time period output by the target logistic regression model. Therefore, the problem of low assessment accuracy of default scores for inclusive enterprises applying for credit products in the related art is solved, and the effect of improving the accuracy of default scores for credit products applied by enterprises is achieved.

[0042] In an exemplary embodiment, determining the target credit data of the target object within a first time period includes: obtaining credit data of multiple dimensions generated by the credit operation performed on the target object within the first time period, wherein the credit data of the multiple dimensions include: financial data, historical credit data, loan data, and macroeconomic data within the first time period; performing preprocessing operations on the credit data of the multiple dimensions to obtain the target credit data, wherein the preprocessing operations include: removing duplicate data in the credit data of each dimension according to credit logic business rules, modifying erroneous data in the credit data of each dimension, and filling missing data in the credit data of each dimension, and the credit logic business rules are determined based on historical credit data.

[0043] Optionally, credit data can be obtained through big data mining technology, financial data includes but is not limited to repayment ability data, debt status data, loan data includes but is not limited to: guarantee status data, and the macroeconomic data in the above first time period includes but is not limited to market data, currency data, and interest rate data. Credit logic business rules are determined based on historical credit data, such as the rationality of the income-to-liability ratio, the matching degree of loan purpose and loan amount, the upper limit of the loan amount, the requirements for repayment period, etc. According to the credit logic business rules, missing values, abnormal values ​​and non-compliant credit data are pre-processed to obtain target credit data, for example: the income information of the target object is missing; there is a target object with a loan amount of 1 billion, while the average loan amount of other objects is about 1 million, and this 1 billion loan amount is obviously an abnormal value; assuming that the loan amount is not allowed to exceed 5 times the target object's income, but the target object's loan amount far exceeds 5 times its income, which exceeds the company's regulations, and this loan amount is obviously not in compliance with the regulations. For these situations, they can be processed according to the credit logic business rules. In this embodiment, the credit data is pre-processed by the credit logic business rules to obtain the target credit data, thereby achieving the purpose of reducing data errors.

[0044] In an exemplary embodiment, constructing a target characteristic variable table of the target credit data includes: performing numerical processing on the target credit data to obtain multiple characteristic variables of the target credit data; combining the multiple characteristic variables according to the correlation between the multiple characteristic variables to derive multiple combined characteristic variables; performing discretization and normalization operations on the multiple characteristic variables to derive multiple transformed characteristic variables; performing crossover operations on the multiple characteristic variables using leaf nodes in a decision tree model to derive multiple crossed characteristic variables; screening the multiple target characteristic variables from the multiple combined characteristic variables, the multiple transformed characteristic variables and the multiple crossed characteristic variables according to screening indicators, wherein the screening indicators include variable missing rate, significance, correlation, group stability index and information value; integrating the multiple target characteristic variables to obtain the target characteristic variable table.

[0045] Optionally, it is assumed that the target credit data specifically includes numerical and non-numerical data such as the target object's income, loan amount, loan time, age, gender, and education level, and after numerical processing, characteristic variables such as loan amount, loan term, target income, target age, target gender, and target education level are obtained. Variables can be combined by the correlation between features. For example, the loan amount and the loan term are multiplied to form a new characteristic variable representing the total loan amount. The numerical characteristic variables are normalized, and the characteristic variables are limited to between 0 and 1. Continuous characteristic variables are converted into discrete variables. For example, age is converted into age groups (youth, middle-aged, and elderly), and income is converted into income levels (low income, medium income, and high income). The leaf nodes in the decision tree model are used to cross the characteristic variables. For example, the loan amount and customer income are cross-operated to obtain a cross characteristic variable representing the relationship between the loan amount and the target income. This embodiment can screen the combined feature variables, the transformed feature variables and the cross feature variables according to the variable missing rate, significance, correlation, group stability index and information value through unsupervised clustering and XGBT (Extreme Gradient Boosting, Chinese name: Extreme Gradient Boosting Tree) algorithm to obtain the target feature variables. This embodiment achieves the purpose of improving the prediction accuracy of the target logistic regression model by deriving and screening related feature variables.

[0046] In an exemplary embodiment, before the target feature variable table is input into the target logistic regression model to obtain the default score of the target object output by the target logistic regression model, the method includes: selecting target sample feature variables from the sample feature variable table according to the marginal contribution of each sample feature variable to obtain the target sample feature variable table; inputting each of the target sample feature variables in the target sample feature variable table into the original logistic regression model one by one, and detecting the output results of the original logistic regression model one by one; when the output results corresponding to each of the target sample feature variables in the target sample feature variable table meet the preset output results, determining that the target logistic regression model is obtained through training.

[0047] In an exemplary embodiment, when the output results corresponding to each of the target sample characteristic variables in the target sample characteristic variable table meet the preset output results, before determining that the target logistic regression model is obtained through training, the method further includes: detecting the sign of the target sample characteristic variable in the target sample characteristic variable table according to a first test rule, wherein the first test rule is determined based on historical default data; calculating the significance score and variance inflation coefficient of the target sample characteristic variable in the target sample characteristic variable table; deleting or replacing the target sample characteristic variables with wrong signs in the target sample characteristic variable table; deleting or replacing the target sample characteristic variables in the target sample characteristic variable table whose significance score is greater than a first threshold; deleting or replacing the target sample characteristic variables in the target sample characteristic variable table whose variance inflation coefficient is greater than a second threshold.

[0048] Optionally, before training the target logistic regression model, the logistic regression results and each target sample characteristic variable need to be tested. For example, if the sign of the target sample characteristic variable does not conform to the provisions of economic theory or the conventions of social and economic practice, the target sample characteristic variable needs to be deleted or replaced regardless of its significance level; to ensure that the final target sample characteristic variables reach a certain significance level; the variance inflation coefficient of the target sample characteristic variable is calculated using the linear regression process. The variance inflation coefficient indicates the degree to which a target sample characteristic variable can be explained by other target sample characteristic variables. The higher the variance inflation coefficient value, the more serious the multicollinearity, and the less accurate the logistic regression result.

[0049] In an exemplary embodiment, when the output results corresponding to each of the target sample feature variables in the target sample feature variable table meet the preset output results, after determining that the target logistic regression model is obtained through training, the method includes: determining a rolling rate of historical default objects, wherein the rolling rate is used to represent the loan change trend of the historical default objects in the historical time period; determining the first default probability of the historical default objects in the historical time period according to the rolling rate; and determining the target default probability of the historical default objects in a third time period based on the first default probability and the output results, wherein the third time period is a time period after the historical time period.

[0050] Optionally, the rolling rate can be calculated by the following method: select a time period, such as a month or a quarter. For each time period, classify the historical default objects according to their overdue status, such as normal, 30 days overdue, 60 days overdue, etc. In the next time period, reclassify the overdue status of the historical default objects. Calculate the change ratio of each overdue status category in the previous and next two time periods. For example, the proportion of historical default objects that change from normal to 30 days overdue, the proportion of historical default objects that recover from 30 days overdue to normal, etc. Add these proportions to get the rolling rate. The rolling rate can analyze the rolling of the overdue status of the historical default objects in different time periods, so as to help adjust and determine the appropriate first default probability according to the rolling trend and proportion. For example, if the rolling rate shows an upward trend, it means that the overdue rate of the historical default objects is low, then the first default probability is low. If the rolling rate shows a downward trend, it means that the default rate of the historical default objects is high, then the first default probability is high. The loan change trend includes but is not limited to the change of the balance after repayment, and the change of the loan that continues to be loaned without repayment. In this embodiment, the overdue status of the historical default object is dynamically analyzed through the rolling rate, so as to achieve the purpose of real-time and effective prediction of the target default score of the target object in the second time period.

[0051] In an exemplary embodiment, determining the target default probability of the historical default object in the third time period based on the first default probability and the output result includes: obtaining a preset score threshold matching the first default probability, wherein the first default probability is inversely proportional to the preset score threshold; comparing the default score with the preset score threshold to obtain a comparison result; determining the target default probability based on the comparison result, wherein, when the comparison result shows that the default score is greater than the preset score threshold, the target default probability is a probability less than the first default probability, and when the comparison result shows that the default score is less than the preset score threshold, the target default probability is a probability greater than the first default probability.

[0052] Optionally, a certain functional relationship can be established between the preset scoring threshold and the first default probability based on expert experience. For example, when the preset scoring threshold is 400, the first default probability is 50% (1:1); for every 40 points that the default score is greater than the preset scoring threshold, the target default probability is 50% less than the first default probability. For example, if the default score is 440, the target default probability is 25%. In this embodiment, the target default probability of the target object in the second time period is determined based on the first default probability and the default score, and the accuracy of the target logistic regression model is trained.

[0053] The present invention will be described below in conjunction with specific embodiments:

[0054] This specific embodiment takes the target logistic regression model created to predict the credit default probability of small and micro enterprises as an example. The process of creating a target logistic regression model is as follows: Figure 3 As shown, the following steps are included:

[0055] S302, mining multi-dimensional credit data such as customer credit and finance: First, conduct in-depth analysis of product customer distribution, data volume, sample time range, etc. in the application scenario of large-amount credit products of more than 5 million yuan under the target logistic regression model to determine the scope of mining sample credit data. Centrally collect multi-dimensional credit data of small and micro enterprises and actual controllers of enterprises;

[0056] S304, cleaning and basic analysis of the mined multi-dimensional credit data: the credit data is cleaned mainly based on logical judgment, business experience and relevant business rules, and the missing values, abnormal values ​​and credit data that do not meet the specifications are analyzed and processed;

[0057] S306, obtain sample data, analyze the definition of good and bad sample data and the window performance period: to ensure the applicability and stability of the target logistic regression model. The selection of sample data needs to combine the credit feature data to select the full amount of data (online credit loan products + offline traditional loan products) for analysis, and select the business sample data close to the current time point as the best, to reduce the deviation of business characteristics caused by the time interval; at the same time, the sample data must also have sufficient performance time (for example, data of more than 12 months). The sample data can be divided into training samples (TRAI N), test samples (VALIDATION) and out-of-time samples (OOT) according to the time sequence. The training samples are used to fit the model parameters, and the test samples and out-of-time samples are used to verify the stability and effectiveness of the model. The appropriateness of the definition of good and bad (corresponding to the above-mentioned first default probability) is evaluated by rolling rate analysis. The rolling rate can analyze the rolling of the customer's overdue status in different time periods, so as to help adjust and determine the appropriate definition of good and bad according to the rolling trend and proportion. The window definition includes an observation window and a performance window. The observation window is the sample application time period selected during the model training process, and usually needs to consider the characteristics of the business's own development and the influence of seasonal effects. The performance window is the performance of the selected modeling group within a given observation window, i.e. the definition of good and bad accounts. A sufficiently long performance period should be selected so that the vast majority of bad accounts will show overdue behavior.

[0058] S308, using the sample data to process feature variables: performing feature crossover, feature combination, data transformation, etc. on the cleaned sample data to generate meaningful feature variables;

[0059] S310, screening variables for model training: In order to improve the efficiency and effect of model training, multiple factors are usually considered comprehensively. Specifically, variable missing rate, significance, correlation, group stability index, information value and expert selection can be used for screening to form a sample feature variable table; in addition, unsupervised clustering and XGBT algorithm can be introduced to select variables;

[0060] S312, the model score result is output by the logistic regression algorithm: the possibility of default of the customer within a given time period is predicted by the regression coefficient value or weight. The logistic regression process of this embodiment adopts the stepwise regression method to select the target sample characteristic variable, select the target sample characteristic variable from the sample characteristic variable table according to the marginal contribution, and check the selected target sample characteristic variables one by one. When all the target sample characteristic variables in the model pass the test and no other variables meet the admission criteria, the stepwise selection procedure ends. For the logistic regression results, the following steps are performed for inspection: Rationality test: The sign of each target sample characteristic variable is checked from the perspective of experience and business rationality. If the sign of the target sample characteristic variable, that is, the trend of its prediction, contradicts the business experience, the variable needs to be deleted or replaced regardless of its significance level. Significance test: For example, according to the results of the Wald chi-square test, ensure that the target sample characteristic variables of the final model reach a certain significance level (p<0.05). Multicollinearity test: For example, the variance inflation factor (VIF) of the model variables is calculated using the linear regression process. VIF indicates the degree to which a target sample characteristic variable can be explained by other target sample characteristic variables. The higher the VIF value, the more serious the multicollinearity. Expert experience exclusion: After the above indicators are met, this embodiment will consider the experience of modeling experts and business experts again, and adjust and optimize the model in combination with the requirements of business access rules. Model verification: This embodiment ensures that the good and bad ratios remain basically consistent when verifying samples and out-of-time samples. When the model is applied to samples at different time points, similar ranking results are produced, indicating that it is a sufficiently robust and effective model.

[0061] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0062] In this embodiment, a device for determining a default score based on a logistic regression model is also provided, and the device is used to implement the above-mentioned embodiments and preferred implementation modes, which have been described and will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0063] Figure 4 is a structural block diagram of a device for determining a default score based on a logistic regression model according to an embodiment of the present application, such as Figure 4 As shown, the device comprises:

[0064] A first determination module 402 is used to determine target credit data of a target object within a first time period, wherein the target object is an inclusive enterprise whose credit line is greater than a preset credit line, and the target credit data includes data generated by credit operations performed by the target object within the first time period;

[0065] A first construction module 404 is used to construct a target characteristic variable table of the target credit data, wherein the target characteristic variable table includes a plurality of target characteristic variables;

[0066] The first input module 406 is used to input the target characteristic variable table into the target logistic regression model to obtain the default score of the target object in the second time period output by the target logistic regression model, where the second time period is a time period after the first time period.

[0067] In an exemplary embodiment, the first determination module 402 includes: a first acquisition submodule, used to acquire credit data of multiple dimensions generated by the credit operation performed on the target object within the first time period, wherein the credit data of multiple dimensions include: financial data, historical credit data, loan data, and macroeconomic data within the first time period; a first execution submodule, used to perform preprocessing operations on the credit data of the multiple dimensions to obtain the target credit data, wherein the preprocessing operations include: removing duplicate data in the credit data of each dimension according to credit logic business rules, modifying erroneous data in the credit data of each dimension, and filling missing data in the credit data of each dimension, and the credit logic business rules are determined based on historical credit data.

[0068] In an exemplary embodiment, the first construction module 404 includes: a first processing submodule, which is used to perform numerical processing on the target credit data to obtain multiple characteristic variables of the target credit data; a first derivation submodule, which is used to combine the multiple characteristic variables according to the correlation between the multiple characteristic variables to derive multiple combined characteristic variables; a second derivation submodule, which is used to perform discretization and normalization operations on the multiple characteristic variables to derive multiple transformed characteristic variables; a third derivation submodule, which is used to perform crossover operations on the multiple characteristic variables using leaf nodes in a decision tree model to derive multiple crossed characteristic variables; a first screening submodule, which is used to screen the multiple target characteristic variables from the multiple combined characteristic variables, the multiple transformed characteristic variables and the multiple crossed characteristic variables according to screening indicators, wherein the screening indicators include variable missing rate, significance, correlation, group stability index and information value; a first integration submodule, which is used to integrate the multiple target characteristic variables to obtain the target characteristic variable table.

[0069] In an exemplary embodiment, the above-mentioned device includes: a first selection module, which is used to input the above-mentioned target feature variable table into the target logistic regression model, and before obtaining the default score of the above-mentioned target object output by the above-mentioned target logistic regression model, select the target sample feature variable from the sample feature variable table according to the marginal contribution of each sample feature variable to obtain the target sample feature variable table; a first detection module, which is used to input each of the above-mentioned target sample feature variables in the above-mentioned target sample feature variable table into the original logistic regression model one by one, and detect the output results of the above-mentioned original logistic regression model one by one; a second determination module, which is used to determine that the above-mentioned target logistic regression model is obtained by training when the output results corresponding to each of the above-mentioned target sample feature variables in the above-mentioned target sample feature variable table meet the preset output results.

[0070] In an exemplary embodiment, the above-mentioned device also includes: a first verification module, which is used to detect the signs of the target sample characteristic variables in the above-mentioned target sample characteristic variable table according to the first verification rule before determining that the output results corresponding to each of the above-mentioned target sample characteristic variables in the above-mentioned target sample characteristic variable table meet the preset output results, wherein the above-mentioned first verification rule is determined based on historical default data; a first calculation module, which is used to calculate the significance score and variance inflation coefficient of the target sample characteristic variables in the above-mentioned target sample characteristic variable table; a first processing module, which is used to delete or replace the target sample characteristic variables with wrong signs in the above-mentioned target sample characteristic variable table; a second processing module, which is used to delete or replace the target sample characteristic variables in the above-mentioned target sample characteristic variable table whose significance scores are greater than the first threshold; and a third processing module, which is used to delete or replace the target sample characteristic variables in the above-mentioned target sample characteristic variable table whose variance inflation coefficients are greater than the second threshold.

[0071] In an exemplary embodiment, the above-mentioned device includes: a third determination module, which is used to determine the rolling rate of the historical default object after determining that the above-mentioned target logistic regression model is obtained through training when the output results corresponding to each of the above-mentioned target sample characteristic variables in the above-mentioned target sample characteristic variable table meet the preset output results, wherein the above-mentioned rolling rate is used to represent the loan change trend of the above-mentioned historical default object in the historical time period; a fourth determination module, which is used to determine the first default probability of the above-mentioned historical default object in the above-mentioned historical time period according to the above-mentioned rolling rate; and a fifth determination module, which is used to determine the target default probability of the above-mentioned historical default object in a third time period based on the above-mentioned first default probability and the above-mentioned output results, wherein the above-mentioned third time period is a time period after the above-mentioned historical time period.

[0072] In an exemplary embodiment, the fifth determination module includes: a second acquisition submodule, used to acquire a preset score threshold matching the first default probability, wherein the first default probability is inversely proportional to the preset score threshold; a first comparison submodule, used to compare the default score with the preset score threshold to obtain a comparison result; and a first determination submodule, used to determine the target default probability based on the comparison result, wherein, when the comparison result shows that the default score is greater than the preset score threshold, the target default probability is a probability less than the first default probability, and when the comparison result shows that the default score is less than the preset score threshold, the target default probability is a probability greater than the first default probability.

[0073] An embodiment of the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps in any one of the above method embodiments are implemented.

[0074] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0075] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0076] An embodiment of the present application further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0077] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0078] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.

[0079] Obviously, those skilled in the art should understand that the above modules or steps of the present application can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that herein, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0080] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for determining a default score based on a logistic regression model, characterized in that: include: Determining target credit data of a target object in a first time period, wherein the target object is an inclusive enterprise whose credit line is greater than a preset credit line, and the target credit data includes data generated by credit operations performed by the target object in the first time period; Constructing a target feature variable table of the target credit data, wherein the target feature variable table includes a plurality of target feature variables; The target characteristic variable table is input into a target logistic regression model to obtain a default score of the target object in a second time period output by the target logistic regression model, wherein the second time period is a time period after the first time period.

2. The method according to claim 1, characterized in that Determine the target credit data of the target object in the first time period, including: Acquire credit data of multiple dimensions generated by the target object performing credit operations within the first time period, wherein the credit data of multiple dimensions include: financial data, historical credit data, loan data, and macroeconomic data within the first time period; Preprocessing operations are performed on the credit data of the multiple dimensions to obtain the target credit data, wherein the preprocessing operations include: removing duplicate data in the credit data of each dimension, modifying erroneous data in the credit data of each dimension, and filling missing data in the credit data of each dimension according to credit logic business rules, and the credit logic business rules are determined based on historical credit data.

3. The method according to claim 1, characterized in that Constructing a target characteristic variable table of the target credit data, including: Performing numerical processing on the target credit data to obtain a plurality of characteristic variables of the target credit data; According to the correlation between the plurality of characteristic variables, the plurality of characteristic variables are combined to derive a plurality of combined characteristic variables; Performing discretization and normalization operations on the plurality of feature variables to derive a plurality of transformed feature variables; Using leaf nodes in a decision tree model to perform a crossover operation on the plurality of feature variables to derive a plurality of crossover feature variables; According to the screening index, a plurality of the target feature variables are screened from the plurality of the combined feature variables, the plurality of the transformed feature variables and the plurality of the cross feature variables to obtain the plurality of the target feature variables, wherein the screening index includes variable missing rate, significance, correlation, population stability index and information value; Integrate a plurality of the target feature variables to obtain the target feature variable table.

4. The method according to claim 1, characterized in that: Before inputting the target characteristic variable table into the target logistic regression model to obtain the default score of the target object output by the target logistic regression model, the method includes: According to the marginal contribution of each sample characteristic variable, a target sample characteristic variable is selected from the sample characteristic variable table to obtain a target sample characteristic variable table; Input each of the target sample characteristic variables in the target sample characteristic variable table into the original logistic regression model one by one, and detect the output results of the original logistic regression model one by one; When the output results corresponding to each of the target sample characteristic variables in the target sample characteristic variable table satisfy the preset output results, it is determined that the target logistic regression model is obtained through training.

5. The method according to claim 1, characterized in that In the case where the output results corresponding to each of the target sample feature variables in the target sample feature variable table satisfy the preset output results, before determining that the target logistic regression model is obtained through training, the method includes: Detecting the sign of the target sample characteristic variable in the target sample characteristic variable table according to a first inspection rule, wherein the first inspection rule is determined according to historical default data; Calculating the significance score and variance inflation coefficient of the target sample characteristic variable in the target sample characteristic variable table; Deleting or replacing target sample feature variables with incorrect signs in the target sample feature variable table; Deleting or replacing the target sample feature variables in the target sample feature variable table whose significance score is greater than the first threshold; Delete or replace the target sample feature variable in the target sample feature variable table whose variance inflation coefficient is greater than the second threshold.

6. The method according to claim 1, characterized in that In the case where the output results corresponding to each of the target sample feature variables in the target sample feature variable table satisfy the preset output results, after determining that the target logistic regression model is obtained through training, the method includes: Determine a rollover rate of a historical default object, wherein the rollover rate is used to represent a loan change trend of the historical default object within a historical time period; Determining the first default probability of the historical default object within the historical time period according to the rolling rate; A target default probability of the historical default object in a third time period is determined based on the first default probability and the output result, wherein the third time period is a time period after the historical time period.

7. The method according to claim 6, characterized in that Determining a target default probability of the historical default object in a third time period based on the first default probability and the output result includes: Obtaining a preset scoring threshold that matches the first default probability, wherein the first default probability is inversely proportional to the preset scoring threshold; Comparing the default score with the preset score threshold to obtain a comparison result; The target default probability is determined based on the comparison result, wherein, when the comparison result shows that the default score is greater than the preset score threshold, the target default probability is a probability less than the first default probability, and when the comparison result shows that the default score is less than the preset score threshold, the target default probability is a probability greater than the first default probability.

8. A device for determining a default score based on a logistic regression model, characterized in that: include: A first determination module is used to determine target credit data of a target object within a first time period, wherein the target object is an inclusive enterprise whose credit line is greater than a preset credit line, and the target credit data includes data generated by credit operations performed by the target object within the first time period; A first construction module is used to construct a target feature variable table of the target credit data, wherein the target feature variable table includes a plurality of target feature variables; The first input module is used to input the target characteristic variable table into the target logistic regression model to obtain the default score of the target object in a second time period output by the target logistic regression model, wherein the second time period is a time period after the first time period.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program implements the steps of the method described in any one of claims 1 to 7 when executed by a processor.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method described in any one of claims 1 to 7 are implemented.

Images