Regression modeling method, device and equipment based on multi-box scheme, medium and product

Abstract

The invention discloses a regression modeling method and device based on a multi-binning scheme, equipment, a medium and a product, and relates to the technical field of financial risk control, and the regression modeling method based on the multi-binning scheme comprises the steps: obtaining each original feature variable in financial customer data, determining at least two binning schemes respectively corresponding to the original characteristic variables and a WOE coding rule of each binning scheme; converting each original characteristic variable based on each WOE coding rule to obtain a plurality of WOE variables corresponding to each original characteristic variable; screening the plurality of WOE variables corresponding to each original characteristic variable to obtain a target WOE variable corresponding to each original characteristic variable; and performing regression modeling according to the target WOE variable of each original feature variable to obtain a target risk control model, each target binning scheme and weight information. According to the method, the constraint of a traditional single binning scheme is broken through, and the risk expression ability of the original feature variables is improved.

Description

Technical Field

[0001] The present application relates to the field of financial risk control technologies, and in particular, to a regression modeling method, device, electronic device, computer-readable storage medium, and computer program product based on a multi-binning scheme. Background Art

[0002] In the field of financial risk control, in order to formulate personalized marketing plans for customers, it is necessary to predict the financial risks of customers. Before predicting the risks of customers, a corresponding risk control model needs to be established. Usually, the risk control model is trained with financial customer data, and before modeling, the feature variables included in the financial customer data need to be binned. The binning scheme will adopt equal-frequency binning or decision tree binning, and the binned feature variables can be used to train the risk control model. However, both equal-frequency binning and decision tree binning can only reflect the risk characteristics in a single dimension and cannot capture the non-linear risk correlation characteristics of variables under different binning schemes, which will lead to data information loss and affect the performance and accuracy of the risk control model.

[0003] The above content is only used to assist in understanding the technical solution of the present application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of the present application is to provide a regression modeling method, device, electronic device, computer-readable storage medium, and computer program product based on a multi-binning scheme, aiming to solve the technical problem that the current binning scheme for variables affects the performance and accuracy of the financial risk control model.

[0005] To achieve the above purpose, the present application proposes a regression modeling method based on a multi-binning scheme. The regression modeling method based on a multi-binning scheme includes:

[0006] Obtain each original feature variable in the financial customer data, and determine at least two binning schemes corresponding to each original feature variable and the WOE coding rules of each binning scheme;

[0007] Convert each original feature variable based on each WOE coding rule to obtain multiple WOE variables corresponding to each original feature variable;

[0008] Screen the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variable corresponding to each original feature variable;

[0009] Perform regression modeling according to the target WOE variables corresponding to each original feature variable to obtain a target risk control model, and the target binning scheme and weight information corresponding to each original feature variable in the target risk control model.

[0010] In one embodiment, the step of determining at least two binning schemes corresponding to each of the original feature variables and the WOE coding rules of each binning scheme includes:

[0011] Generate an equal-frequency binning scheme and a decision tree binning scheme respectively corresponding to each of the original feature variables, wherein the equal-frequency binning scheme includes sub-schemes with multiple different quantile values;

[0012] Obtain the WOE coding rules respectively corresponding to each of the equal-frequency binning schemes and the decision tree binning schemes.

[0013] In one embodiment, after the step of converting each of the original feature variables based on each of the WOE coding rules to obtain multiple WOE variables respectively corresponding to each of the original feature variables, the method further includes:

[0014] Name each of the WOE variables based on the name of each original feature vector and the attributes of each binning scheme, wherein the attributes include type and / or quantile value;

[0015] Construct a multi-dimensional binned WOE variable pool based on the named WOE variables, wherein the multi-dimensional binned WOE variable pool is used for logistic regression modeling and variable screening.

[0016] In one embodiment, the step of screening the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variable respectively corresponding to each original feature variable includes:

[0017] Initialize an empty initial regression model;

[0018] Calculate the significance of each WOE variable, and sequentially add the WOE variables with significance higher than the preset addition significance threshold to the initial regression model; and / or,

[0019] Calculate the significance of the WOE variables already added to the initial regression model, and sequentially remove the WOE variables with significance lower than the preset removal significance threshold;

[0020] When there is no WOE variable with significance higher than the addition significance threshold among the WOE variables not added to the model, and there is no WOE variable with significance lower than the removal significance threshold among the WOE variables in the model, determine the WOE variables in the initial regression model as the target WOE variables, wherein the target WOE variables correspond to the original feature variables one by one.

[0021] In one embodiment, the step of screening the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variable respectively corresponding to each original feature variable includes:

[0022] Initialize the LASSO model and the objective function corresponding to the LASSO model, where the objective function includes a regularization parameter and regression coefficients corresponding to each WOE variable.

[0023] Train the LASSO model to determine the values of the regression coefficients corresponding to each WOE variable.

[0024] Determine the WOE variable with the largest regression coefficient among the WOE variables corresponding to each original feature variable as the target WOE variable, where the target WOE variable corresponds one-to-one with the original feature variable.

[0025] In one embodiment, the step of performing regression modeling according to the target WOE variables corresponding to each of the original feature variables to obtain a target risk control model, and the target binning scheme and weight information corresponding to each original feature variable in the target risk control model includes:

[0026] Perform logistic regression modeling according to each of the target WOE variables to obtain a first risk control model.

[0027] Evaluate the first risk control model through preset verification sample data to obtain the AUC value and KS value corresponding to the first risk control model.

[0028] If one of the AUC value and KS value does not meet the corresponding preset threshold, return to execute the step: screen the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variables corresponding to each of the original feature variables.

[0029] If both the AUC value and KS value meet the corresponding preset thresholds, determine the first risk control model as the target risk control model.

[0030] Determine the target binning scheme corresponding to each of the original feature variables based on the names of the target WOE variables corresponding to each of the original feature variables.

[0031] Obtain the weight information corresponding to each original feature variable in the target risk control model.

[0032] In addition, to achieve the above object, the present application also proposes a regression modeling device based on a multi-binning scheme. The regression modeling device based on a multi-binning scheme includes:

[0033] A scheme determination module, configured to obtain each original feature variable in the financial customer data, and determine at least two binning schemes corresponding to each of the original feature variables and the WOE coding rules of each of the binning schemes.

[0034] A variable encoding module, configured to convert each of the original feature variables based on each of the WOE encoding rules, so as to obtain a plurality of WOE variables respectively corresponding to each of the original feature variables;

[0035] A variable screening module, configured to screen the plurality of WOE variables respectively corresponding to each original feature variable, so as to obtain the target WOE variables respectively corresponding to each of the original feature variables;

[0036] A model building module, configured to perform regression modeling according to the target WOE variables respectively corresponding to each of the original feature variables, so as to obtain a target risk control model, and the target binning scheme and weight information respectively corresponding to each original feature variable in the target risk control model.

[0037] In addition, to achieve the above object, the present application further provides an electronic device, the device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program is configured to implement the steps of the regression modeling method based on a multi-binning scheme as described above.

[0038] In addition, to achieve the above object, the present application further provides a storage medium, the storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium, and when the computer program is executed by a processor, the steps of the regression modeling method based on a multi-binning scheme as described above are implemented.

[0039] In addition, to achieve the above object, the present application further provides a computer program product, the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the regression modeling method based on a multi-binning scheme as described above are implemented.

[0040] The present application proposes a regression modeling method based on a multi-binning scheme. In the regression modeling method based on the multi-binning scheme, first, each original feature variable in the financial customer data is obtained, and at least two binning schemes corresponding to each of the original feature variables and the WOE coding rules of each of the binning schemes are determined. Then, each of the original feature variables is transformed based on each of the WOE coding rules to obtain a plurality of WOE variables corresponding to each of the original feature variables. Next, the plurality of WOE variables corresponding to each original feature variable are screened to obtain the target WOE variables corresponding to each of the original feature variables. Finally, regression modeling is performed based on the target WOE variables corresponding to each of the original feature variables to obtain a target risk control model, as well as the target binning scheme and weight information corresponding to each original feature variable in the target risk control model. In the technical solution of the present application, when binning each original feature variable, the WOE coding rules corresponding to multiple binning schemes are adopted to bin and transform each original feature variable, so that the multiple WOE variables corresponding to a single original feature variable can express the corresponding risk characteristics in multiple dimensions, increasing the dimension of the feature work. Subsequently, screening is performed again, and the selected original feature variables are relatively the best target WOE variables. In this way, among the variables for constructing the target risk control model, there are WOE variables obtained by various binning methods. Compared with the traditional single binning scheme, the multi-dimensional WOE variables can comprehensively reflect the non-linear risk correlation characteristics of various variables in different segments, with less loss of data information. The synergistic effect among the multiple variables of various binning schemes can enable the target risk control model to be limited by the single binning scheme, with richer feature dimensions, better prediction performance and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present application and, together with the specification, are used to explain the principles of the present application.

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0043] Figure 1 It is a schematic flowchart provided for Embodiment 1 of the regression modeling method based on the multi-binning scheme of the present application;

[0044] Figure 2 It is a schematic flowchart of the traditional regression modeling process based on single-variable binning;

[0045] Figure 3 It is a schematic flowchart of the regression modeling process based on single-variable parallel binning in the embodiment of the present application

[0046] Figure 4 It is a schematic structural diagram of the regression modeling device based on the multi-binning scheme in the embodiment of the present application;

[0047] Figure 5 It is a schematic diagram of the device structure of the hardware operating environment involved in the regression modeling method based on the multi-binning scheme in the embodiment of the present application.

[0048] The implementation, functional features, and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. Specific Embodiments

[0049] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0050] To better understand the technical solutions of the present application, the following will be described in detail with reference to the accompanying drawings of the specification and specific embodiments.

[0051] The execution subject of this embodiment can be a computing service device with data processing, network communication, and program running functions, such as a tablet computer, a personal computer, a mobile phone, a server, etc., or an electronic device, a control device, etc. that can implement the above functions. Hereinafter, taking a computer as the execution subject as an example, this embodiment and the following embodiments will be described.

[0052] The embodiment of the present application provides a regression modeling method based on the multi-binning scheme. Refer to Figure 1 , Figure 1 It is a schematic flowchart of the first embodiment of the regression modeling method based on the multi-binning scheme of the present application. The regression modeling method based on the multi-binning scheme applied to the local end includes:

[0053] Step S10, obtain each original feature variable in the financial customer data, and determine at least two binning schemes corresponding to each original feature variable and the WOE coding rules of each binning scheme;

[0054] The regression modeling method based on the multi-binning scheme of the embodiment of the present application is applied in the financial field. Among them, the financial customer data is customer data from financial institutions such as banks, including information data in various dimensions such as the customer's fund information, identity information, consumption information, and behavior information.

[0055] Among them, the original feature variable refers to the specific type of data in the financial customer data. For example, the customer's age, annual income, consumption type, loan amount, etc. all belong to the original feature variables. Before performing regression modeling, it is necessary to bin the original feature variables to capture the correlation between the features and risks, and at the same time improve the stability and interpretability.

[0056] In another feasible embodiment, after the binning scheme for each original feature variable is determined, binning verification is required to ensure that the sample size in each bin is sufficient (e.g., ≥30), and the distribution trend of good and bad samples conforms to business logic (e.g., the default rate increases monotonically). Outlier processing can also be performed: set extreme values as independent bins separately to avoid interfering with the model stability.

[0057] Step S20, convert each original feature variable based on each WOE coding rule to obtain multiple WOE variables respectively corresponding to each original feature variable;

[0058] Perform conversion processing on each original feature variable according to different binning schemes. For example, dummy variable conversion: convert the binned categorical variable into 0 / 1 coding (such as generating labels like "<25 years old = 0" after age binning). Exemplarily, WOE (Weight of Evidence) is used to measure the relative risk of good and bad samples in a bin, and the formula is: WOE = ln(good sample ratio / bad sample ratio). Moreover, a single original feature variable can obtain multiple WOE variables according to the coding rules of multiple binning schemes.

[0059] Compared with the traditional single binning scheme, the binning scheme of the embodiment of the present application can obtain richer and more dimensional WOE variables after binning, and can capture the characteristics of WOE variables in different segments. In practical scenarios, generating derivative variables through multiple binning schemes can enable a single original variable to express different risk patterns according to different segments, increasing the feature engineering dimension by 5 - 10 times. Moreover, while increasing the feature dimension and improving the model performance, the interpretability of the risk control model is maintained by using the WOE coding mechanism.

[0060] Step S30, screen the multiple WOE variables respectively corresponding to each original feature variable to obtain the target WOE variable respectively corresponding to each original feature variable;

[0061] After obtaining multiple WOE variables corresponding to each original feature variable, certain screening is required to obtain a target WOE variable respectively corresponding to each original feature variable. The purpose of this step is to screen out the best target WOE variable among the WOE variables corresponding to each original feature variable to avoid duplicate data of a single original feature variable during the modeling process.

[0062] It can be understood that each original feature variable corresponds to multiple WOE variables, and there can be multiple different variable combinations. The purpose of screening is to obtain an optimal variable combination, and finally a variable combination is jointly constituted by the screened target WOE variables. Moreover, the two-stage architecture of pre-generation of the binning scheme and dynamic screening of the variable combination is adopted in the embodiment of the present application, reducing the computational complexity while ensuring the model optimization effect.

[0063] Exemplarily, during the process of variable screening, one can refer to the traditional screening scheme for single-binned variables, perform screening based on the stepwise regression method or the LASSO (Least Absolute Shrinkage and Selection Operator) regression method, and combine the variance inflation factor (VIF) test for multicollinearity to examine the correlation between the variables included in the model. In regression analysis, multicollinearity refers to a high degree of correlation between two or more independent variables (i.e., WOE variables) in the model, resulting in unstable model estimation results and coefficients that are difficult to interpret. The VIF is the core tool for detecting collinearity because high collinearity leads to an increase in the variance of coefficient estimates, making hypothesis testing unreliable (such as the failure of the t-test). For example, if two variables are highly correlated, the model may randomly assign a significant coefficient to one variable and an insignificant coefficient to the other. Under collinearity, the sign of the coefficient may contradict the theoretical expectation (such as a negative correlation between "years of education" and "income"). By deleting variables with high VIF or combining related variables, the business interpretability of the coefficients can be restored. At the initial stage of model building, the VIF is used to guide the variable screening process of stepwise regression or regularization (such as the LASSO regression method).

[0064] Compared with the traditional regression modeling scheme, there are differences in four comparison dimensions in the embodiments of the present application. Firstly, in terms of the generation method of the binning scheme, the traditional one is a single-threaded fixed rule, while in the present application, multiple rules (i.e., sub-case schemes) are generated in parallel. Secondly, in terms of the WOE variable dimension, in the traditional scheme, the number of WOE variables is the same as that of the original feature variables, both being P, while in the present application, it is extended through a multi-binning scheme, and the number of extended variables = P × M (M is the number of binning schemes). Thirdly, in terms of the variable screening mechanism, in the traditional scheme, variable screening is performed once, that is, only the variables corresponding to a single binning scheme are screened, while in the present application, it also includes screening the combinations of WOE variables corresponding to each binning scheme. Fourthly, in terms of the upper limit of model performance, the upper limit of the model performance in the traditional scheme is limited by the selected initial binning scheme, while in the present application, multiple binning schemes are initially adopted and then variable screening is performed, breaking through the limitations of a single binning scheme.

[0065] Step S40: Perform regression modeling based on the target WOE variables corresponding to each original feature variable to obtain a target risk control model, as well as the target binning scheme and weight information corresponding to each original feature variable in the target risk control model.

[0066] After obtaining the target WOE variable corresponding to each original feature variable, further perform regression modeling. The methods of regression modeling can adopt the stepwise regression method, the LASSO regression method, etc. commonly used in traditional modeling schemes to obtain a target risk control model established through logistic regression. This target risk control model can predict the financial risk situation of customers based on the input variables.

[0067] In addition, the target binning scheme corresponding to each original feature variable in the target risk control model can be determined according to the type of the target WOE variable determined as described above. Different types of WOE variables correspond to different binning schemes. The weight information represents the weights corresponding to various original feature variables in the target risk control model and can be directly read from the target risk control model. The weight information is gradually determined during the modeling and training process of the target risk control model.

[0068] After the target risk control model is established, the target binning scheme and weight information corresponding to each original feature variable can provide guidance in the subsequent process of updating the target risk control model. In this way, there is no need to build the model from scratch. The target binning scheme of these original feature variables can be used as the data basis, and it can also provide a reference for the establishment of other similar risk control models, improving the modeling efficiency of subsequent risk control models and the performance and accuracy of the risk control models.

[0069] The embodiment of the present application proposes a regression modeling method based on a multi-binning scheme. In the regression modeling method based on a multi-binning scheme, first, each original feature variable in the financial customer data is obtained, at least two binning schemes corresponding to each of the original feature variables and the WOE coding rules of each of the binning schemes are determined. Then, each of the original feature variables is transformed based on each of the WOE coding rules to obtain multiple WOE variables corresponding to each of the original feature variables. Then, the multiple WOE variables corresponding to each original feature variable are screened to obtain the target WOE variable corresponding to each of the original feature variables. Finally, regression modeling is performed according to the target WOE variables corresponding to each of the original feature variables to obtain the target risk control model, as well as the target binning scheme and weight information corresponding to each original feature variable in the target risk control model. In the technical solution of the embodiment of the present application, when binning each original feature variable, the WOE coding rules of multiple binning schemes are adopted to bin and transform each original feature variable, so that the multiple WOE variables corresponding to a single original feature variable can express the corresponding risk characteristics in multiple dimensions, increasing the dimension of the feature work. Then, screening is performed to screen out the relatively best target WOE variable of the original feature variable. In this way, among the variables for constructing the target risk control model, there are WOE variables obtained by various binning methods. Compared with the traditional single binning scheme, the multi-dimensional WOE variables can comprehensively reflect the non-linear risk correlation characteristics of various variables in different segments, with less data information loss. The synergistic effect among multiple variables of various binning schemes can enable the target risk control model to be free from the limitations of a single binning scheme, with richer feature dimensions, better prediction performance and accuracy.

[0070] Further, in a feasible implementation manner, the step of determining at least two binning schemes corresponding to each original feature variable and the WOE coding rules of each of the binning schemes may include:

[0071] Step S11: Generate an equal-frequency binning scheme and a decision-tree binning scheme respectively corresponding to each original feature variable, where the equal-frequency binning scheme includes sub-schemes with multiple different quantile values.

[0072] Step S12: Obtain the WOE encoding rules respectively corresponding to each equal-frequency binning scheme and decision-tree binning scheme.

[0073] For multiple original feature vectors, differential binning schemes can be generated. Among them, the equal-frequency binning scheme can specifically include sub-schemes with multiple different division point values, such as 3%, 5%, 10%, while the decision-tree binning scheme can perform binning based on different depth difference criteria of the decision tree.

[0074] Exemplarily, the equal-frequency binning scheme is applicable to continuous variables and rough classification of discrete variables. Among them, equal-frequency binning divides continuous variables into several intervals equally according to the sample size, and is applicable to scenarios where the data distribution is uneven (such as income, age, etc.). For example, 10,000 age data are evenly divided into 4 bins, with 2,500 people in each bin, ensuring the stability of the sample size, reducing the interference of outliers, and improving the stability of the model. In addition, when the number of categories of discrete variables (such as industries, cities) is too large, the low-frequency categories can be merged through equal-frequency binning to reduce the dimension. For example, 100 industries are merged into 10 categories.

[0075] The decision-tree binning scheme uses continuous variables and discrete variables in high-cardinality scenarios. Decision-tree binning selects the optimal splitting point through the target variable (such as the default label), and is applicable to capturing non-linear relationships. For example, use CART (Classification and Regression Trees) to bin the income, automatically identify the risk inflection point, can effectively combine the target variable, and the binning result is more in line with the business logic. For high-cardinality discrete variables (such as addresses, device models), decision-tree binning can merge low-frequency categories to reduce the sparse matrix problem. For example, 1,000 cities are merged into 10 regions to improve the generalization ability of the model and avoid overfitting.

[0076] In addition, during the binning process, it is necessary to determine the corresponding WOE (Weight of Evidence) coding rules. The purpose of WOE coding is to encode various forms of original features, facilitating subsequent further data processing. The WOE coding rules can be determined according to the actual situation of the original feature vector and are used to bin variables (equal frequency, equal width, or binning based on decision trees). Specifically, the WOE value is calculated for each bin respectively, and the bin WOE is used to replace the original value. WOE coding solves the non-linear problem of continuous variables and improves the robustness of the model. The WOE coding rules can be customized according to the form of the original feature variables. For example, for numerical variables, through a fixed calculation formula, and for non-numerical (such as categorical) variables, they can be transformed through predefined conversion rules.

[0077] Further, after the step of converting each of the original feature variables based on the respective WOE coding rules to obtain multiple WOE variables corresponding to each of the original feature variables, the method may further include:

[0078] Step S13, naming each of the WOE variables based on the name of each original feature vector and the attributes of each of the binning schemes, where the attributes include type and / or quantile value;

[0079] Step S14, constructing a multi-dimensional binned WOE variable pool based on the named WOE variables, where the multi-dimensional binned WOE variable pool is used for logistic regression modeling and variable screening.

[0080] To make the management of WOE variables more intuitive and standardized, a corresponding multi-dimensional binned WOE variable pool is generated based on the determined WOE variables corresponding to each original feature variable.

[0081] For example, if the name of the original feature vector is X1, the type of the corresponding binning scheme is equal-frequency binning (corresponding to P), and the quantile value is 3%, it is named X1_WOE_P3; if X1 is binned according to the binning scheme with a 5% quantile, the corresponding name is X1_WOE_P5, and so on.

[0082] Construct the named WOE variables into a multi-dimensional binned WOE variable pool for subsequent logistic regression modeling and variable screening.

[0083] In a feasible embodiment, the step of screening the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variable corresponding to each original feature variable may include:

[0084] Step S31, initializing an empty initial regression model;

[0085] Step S32: Calculate the significance of each WOE variable, and successively add the WOE variables with significance higher than the preset addition significance threshold to the initial regression model; and / or,

[0086] Step S33: Calculate the significance of the WOE variables already added to the initial regression model, and successively remove the WOE variables with significance lower than the preset removal significance threshold;

[0087] Step S34: When there is no WOE variable with significance higher than the addition significance threshold among the WOE variables not added to the model, and there is no WOE variable with significance lower than the removal significance threshold among the WOE variables in the model, determine the WOE variables in the initial regression model as the target WOE variables, where the target WOE variables correspond one-to-one with the original feature variables.

[0088] The embodiment of the present application provides a method for screening WOE variables by the stepwise regression method. It should be noted that the initial regression model is constructed first: starting from an empty model (or a full variable model), the addition significance threshold and the removal significance threshold are set.

[0089] Specifically, in the process of variable screening: it can include an addition process and a removal process. The addition process includes: calculating the F statistic or AIC (Akaike information criterion, which measures the goodness of fit of a statistical model) value of each candidate variable, and selecting the variables with significance higher than the addition significance threshold to be added to the model, where the F statistic or AIC value is used to reflect the significance of the WOE variable, and the F statistic is a statistical index for evaluating the significance of a single variable to the model. Its core is to judge whether the variable has significant explanatory power for the target variable through analysis of variance.

[0090] The removal process includes: performing a significance test (such as a t test) on each WOE variable added to the initial regression model, calculating its F statistic or AIC value, and removing the WOE variables with significance lower than the removal significance threshold from the model, where the addition significance threshold and the removal significance threshold are not equal and can be set according to the actual situation. For example, the addition significance threshold can be greater than the removal significance threshold. Before and after the WOE variable is added to the model, its significance will change and will also be affected by other added or removed WOE variables.

[0091] Iterative optimization includes: repeating the above processes of introducing and removing WOE variables until no WOE variable can be added or removed. It should be noted that the WOE variables finally retained in the initial regression model need to meet the constraint condition that only one corresponding WOE variable is retained for the same original feature variable value. In the stepwise regression process, it can only include the step of gradually adding (forward regression), or only include the step of gradually removing (backward regression), or both can be included.

[0092] In another feasible embodiment, the step of screening the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variable corresponding to each original feature variable may include:

[0093] Step S35, initialize the LASSO model and the objective function corresponding to the LASSO model, where the objective function includes a regularization parameter and the regression coefficients corresponding to each WOE variable;

[0094] Step S36, train the LASSO model to determine the values of the regression coefficients corresponding to each WOE variable;

[0095] Step S37, determine the WOE variable with the largest regression coefficient among the WOE variables corresponding to each original feature variable as the target WOE variable, where the target WOE variable corresponds one-to-one with the original feature variable.

[0096] The embodiment of the present application also provides another method for variable screening based on the LASSO model. First, initialize a LASSO model and an objective function.

[0097] Exemplarily, the goal of LASSO regression is to minimize the following objective function:

[0098]

[0099] Where X and Y are the input variables and the output variable respectively, N is the number of samples, β is the regression coefficient vector, λ is the regularization parameter used to control the model complexity, and ||β||1 is the L1 norm (sum of absolute values) of the regression coefficients.

[0100] The choice of the regularization parameter, the magnitude of λ determines the sparsity of the model: when λ is closer to 0, the model degenerates into ordinary linear regression; when λ is closer to positive infinity, all regression coefficients are compressed to zero. Usually, the optimal λ is selected through cross-validation (such as k-fold cross-validation) to make the performance of the model on the validation set the best.

[0101] During the process of training the LASSO model, an optimization algorithm (such as the coordinate descent method or the gradient descent method) is used to solve the LASSO regression problem to obtain the regression coefficient β. During the solution process, the L1 regularization will compress the coefficients of some unimportant variables to zero, thereby realizing variable screening. Use cross-validation to evaluate the model performance under different λ, and select the λ that makes the performance of the validation set the best.

[0102] Specifically, according to the trained LASSO model, check the regression coefficient β, and retain those WOE variables with non-zero coefficients. These variables can be determined as important explanatory variables. For the binned WOE variables, if the regression coefficient of a certain WOE variable is zero, it indicates that the explanatory power of this variable for the dependent variable is weak. If the regression coefficients of a certain WOE variable are all zero, it indicates that this variable has no significant impact on the dependent variable as a whole. Therefore, selecting the WOE variable with the relatively largest regression coefficient as the target WOE variable corresponding to the original feature variable can effectively improve the contribution degree of the variable and ensure the performance and prediction accuracy of the finally trained risk control model.

[0103] Further, in a feasible embodiment, the step of performing regression modeling according to the target WOE variables respectively corresponding to the original feature variables to obtain a target risk control model, and the target binning scheme and weight information respectively corresponding to the original feature variables in the target risk control model includes:

[0104] Step S41, perform logistic regression modeling according to each target WOE variable to obtain a first risk control model;

[0105] As described above, the embodiments of the present application provide two different methods for screening WOE variables. After screening the WOE variables, complete the logistic regression modeling following the aforementioned screening methods. For example, in the process of screening WOE variables by the stepwise regression method, perform forward stepwise regression and backward stepwise regression respectively. When there are no adjustable WOE variables, the corresponding first risk control model can be obtained. When screening WOE variables based on the LASSO model, while training the LASSO model, perform variable screening through the regression coefficients of each WOE variable. After the screening is completed, the first risk control model can be obtained.

[0106] Step S42, perform model evaluation on the first risk control model through preset verification sample data to obtain the AUC value and KS value corresponding to the first risk control model;

[0107] To ensure the performance of the model, out-of-sample validation or cross-validation is required, and the performance of the first risk control model is evaluated through indicators such as AUC (Area Under the ROC Curve, the area enclosed by the ROC curve and the coordinate axes) and KS (Kolmogorov-Smirnov Statistic). AUC is the area under the ROC curve. The ROC curve takes the false positive rate as the horizontal axis and the true positive rate as the vertical axis, and is drawn by adjusting the classification threshold. The closer the AUC value is to 1, the stronger the ability of the model to distinguish positive and negative classes. The KS value measures the maximum vertical distance between the cumulative distribution functions of positive and negative classes, reflecting the discrimination ability of the model at the optimal threshold. The AUC and KS values are two core indicators for evaluating the model performance, measuring the model from the perspectives of overall discrimination ability and discrimination at the best threshold respectively.

[0108] Step S43, if one of the AUC value and the KS value does not meet the corresponding preset threshold, return to execute Step S30: screen the multiple WOE variables corresponding to each original feature variable respectively to obtain the target WOE variables corresponding to each original feature variable;

[0109] Step S44, if both the AUC value and the KS value meet the corresponding preset thresholds, determine the first risk control model as the target risk control model;

[0110] Only when both the AUC value and the KS value of the first risk control model meet the corresponding preset thresholds can it be determined that the performance of the first risk control model meets the requirements. Otherwise, return to the step of re-screening WOE variables until the performance of the obtained first risk control model meets the requirements.

[0111] Step S45, based on the names of the target WOE variables corresponding to each original feature variable respectively, determine the target binning scheme corresponding to each original feature variable;

[0112] When outputting the target binning scheme corresponding to each original feature variable respectively, it can be presented in the form of a WOE mapping table. For example, the WOE mapping table reflects the WOE coding rules and binning schemes corresponding to various original feature variables respectively.

[0113] Step S46, obtain the weight information corresponding to each original feature variable in the target risk control model.

[0114] The weight information corresponding to each original feature variable in the target risk control model expresses the influence degree of each original feature variable on the dependent variable of the target risk control model. The greater the weight, the higher the importance of the original feature variable and the greater the impact on the final risk assessment result of the customer.

[0115] For the convenience of understanding, a detailed comparison is made between the regression modeling processes of the traditional approach and the present application. The traditional regression modeling process based on univariate binning is as follows Figure 2 As described, first, data preprocessing is performed, including processes such as data inspection / sample splitting / missing value handling, etc. Then, a unique (binning) scheme is determined through univariate binning, and the original feature variables (X1, X2... X p ) are binned to obtain the WOE variables X1_woe, X2_woe,... X p _woe after WOE transformation. After that, multivariate analysis is performed, including methods such as Stepwise (stepwise regression) / Lasso / VIF for variable screening to construct the corresponding risk control model, and the model evaluation is completed by calculating values such as KS / AUC of the risk control model. The regression modeling process based on univariate parallel binning in the embodiments of the present application is as follows Figure 3 As shown, first, data preprocessing is performed, including processes such as data inspection / sample splitting / missing value handling, etc. Then, the original feature variables (X1, X2... X p ) are binned respectively through binning scheme 1 - binning scheme 20, where the binning schemes include at least the 1% quantile, 2% quantile... 20% quantile, etc. After that, multivariate analysis is performed, including methods such as Stepwise (stepwise regression) / Lasso / VIF for variable screening to construct the corresponding risk control model, and the model evaluation is completed by calculating values such as KS / AUC of the risk control model. Through practical tests, in the embodiments of the present application, by adopting variable binning with multiple binning schemes and an innovative dynamic optimization mechanism, the preferred binning scheme link is postponed from univariate analysis to the multivariate analysis stage, and the optimization effect is improved by utilizing the synergistic effect between variables. Experimental data shows that the AUC value is increased by 2% - 5%.

[0116] It should be noted that the above examples are only for understanding the present application and do not constitute a limitation on the regression modeling method based on multiple binning schemes of the present application. Any simple transformation in more forms based on this technical concept is within the protection scope of the present application.

[0117] The present application also provides a regression modeling device based on multiple binning schemes. Referring to Figure 4 , the regression modeling device based on multiple binning schemes includes:

[0118] A scheme determination module 10, configured to obtain each original feature variable in the financial customer data, and determine at least two binning schemes corresponding to each of the original feature variables and the WOE coding rules of each of the binning schemes;

[0119] A variable coding module 20, configured to convert each of the original feature variables based on each of the WOE coding rules to obtain multiple WOE variables corresponding to each of the original feature variables;

[0120] A variable screening module 30, configured to screen multiple WOE variables corresponding to each original feature variable respectively, so as to obtain target WOE variables corresponding to each of the original feature variables;

[0121] A model building module 40, configured to perform regression modeling according to the target WOE variables corresponding to each of the original feature variables respectively, so as to obtain a target risk control model, and a target binning scheme and weight information corresponding to each original feature variable in the target risk control model.

[0122] In one embodiment, the scheme determination module 10 is further configured to:

[0123] Generate an equal-frequency binning scheme and a decision tree binning scheme corresponding to each of the original feature variables respectively, wherein the equal-frequency binning scheme includes sub-schemes with multiple different quantile values;

[0124] Obtain WOE coding rules corresponding to each of the equal-frequency binning schemes and decision tree binning schemes respectively.

[0125] In one embodiment, the regression modeling device based on multiple binning schemes further includes a variable pool construction module, and the variable pool construction module is configured to:

[0126] Name each of the WOE variables based on the name of each original feature vector and the attributes of each of the binning schemes, wherein the attributes include type and / or quantile value;

[0127] Construct a multi-dimensional binned WOE variable pool based on each of the named WOE variables, wherein the multi-dimensional binned WOE variable pool is used for logistic regression modeling and variable screening.

[0128] In one embodiment, the variable screening module 30 is further configured to:

[0129] Initialize an empty initial regression model;

[0130] Calculate the significance of each WOE variable, and sequentially add WOE variables with significance higher than a preset addition significance threshold to the initial regression model; and / or,

[0131] Calculate the significance of the WOE variables that have been added to the initial regression model, and sequentially remove WOE variables with significance lower than a preset removal significance threshold;

[0132] When there is no WOE variable with significance higher than the addition significance threshold among the WOE variables not added to the model, and there is no WOE variable with significance lower than the removal significance threshold among the WOE variables in the model, determine the WOE variables in the initial regression model as target WOE variables, wherein the target WOE variables correspond one-to-one with the original feature variables.

[0133] In one embodiment, the variable screening module 30 is further configured to:

[0134] Initialize the LASSO model and the objective function corresponding to the LASSO model, where the objective function includes a regularization parameter and regression coefficients corresponding to each WOE variable;

[0135] Train the LASSO model to determine the values of the regression coefficients corresponding to each WOE variable;

[0136] Determine the WOE variable with the largest regression coefficient among the WOE variables corresponding to each original feature variable as the target WOE variable, where the target WOE variable corresponds one-to-one with the original feature variable.

[0137] In one embodiment, the model establishment module 40 is further configured to:

[0138] Perform logistic regression modeling based on each of the target WOE variables to obtain a first risk control model;

[0139] Evaluate the first risk control model through preset verification sample data to obtain the AUC value and KS value corresponding to the first risk control model;

[0140] If one of the AUC value and the KS value does not meet the corresponding preset threshold, return to execute the step: screen the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variable corresponding to each original feature variable;

[0141] If both the AUC value and the KS value meet the corresponding preset thresholds, determine the first risk control model as the target risk control model;

[0142] Based on the names of the target WOE variables corresponding to each original feature variable, determine the target binning scheme corresponding to each original feature variable;

[0143] Obtain the weight information corresponding to each original feature variable in the target risk control model.

[0144] The regression modeling device based on a multi-binning scheme provided by the present application adopts the regression modeling method based on a multi-binning scheme in the above embodiment, and can solve the technical problem that the current binning scheme for variables affects the performance and accuracy of the financial risk control model. Compared with the prior art, the beneficial effects of the regression modeling device based on a multi-binning scheme provided by the present application are the same as those of the regression modeling method based on a multi-binning scheme provided by the above embodiment, and other technical features in the regression modeling device based on a multi-binning scheme are the same as the features disclosed in the method of the previous embodiment, and will not be elaborated here.

[0145] The present application provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the regression modeling method based on the multi-bin scheme in the first embodiment above.

[0146] Reference is made below to Figure 5 , which shows a schematic structural diagram of an electronic device suitable for implementing the embodiments of the present application. The electronic device in the embodiments of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions: tablet computers), PMPs (Portable Media Players), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The electronic device shown is only an example and should not impose any limitation on the functions and usage scope of the embodiments of the present application.

[0147] As Figure 5As shown, the electronic device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which may perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM: Random Access Memory) 1004. In the RAM 1004, various programs and data required for the operation of the electronic device are also stored. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems may be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 may allow the electronic device to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows an electronic device with various systems, it should be understood that it is not required to implement or have all the shown systems. Instead, more or fewer systems may be implemented or had.

[0148] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from a network through the communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the methods of the embodiments disclosed in the present application are executed.

[0149] The electronic device provided in the present application adopts the regression modeling method based on the multi-binning scheme in the above embodiments, and can solve the technical problem that the current binning scheme for variables affects the performance and accuracy of the financial risk control model. Compared with the prior art, the beneficial effects of the electronic device provided in the present application are the same as those of the regression modeling method based on the multi-binning scheme provided in the above embodiments, and other technical features in the electronic device are the same as those disclosed in the method of the previous embodiment, and will not be elaborated herein.

[0150] It should be understood that each part disclosed in this application can be implemented by hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0151] As described above, the above are only specific embodiments of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0152] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the regression modeling method based on the multi-binning scheme in the above embodiments.

[0153] The computer-readable storage medium provided by this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, device, or device. The program code contained on the computer-readable storage medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0154] The above computer-readable storage medium can be included in an electronic device; it can also exist separately without being assembled into the electronic device.

[0155] The above computer-readable storage medium carries one or more programs, which, when executed by an electronic device, cause the electronic device to perform the following: obtaining each original feature variable in financial customer data, determining at least two binning schemes corresponding to each of the original feature variables and the WOE coding rules of each binning scheme; converting each of the original feature variables based on each of the WOE coding rules to obtain a plurality of WOE variables corresponding to each of the original feature variables; screening the plurality of WOE variables corresponding to each original feature variable to obtain a target WOE variable corresponding to each of the original feature variables; performing regression modeling based on the target WOE variables corresponding to each of the original feature variables to obtain a target risk control model, and the target binning scheme and weight information corresponding to each original feature variable in the target risk control model.

[0156] Computer program code for performing the operations of the present application may be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may execute entirely on the user's computer, partially on the user's computer, execute as a stand-alone software package, execute partially on the user's computer and partially on a remote computer, or execute entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., by connecting through the Internet using an Internet service provider).

[0157] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, may be implemented by a dedicated hardware-based system for performing the specified functions or operations, or may be implemented by a combination of dedicated hardware and computer instructions.

[0158] The modules involved in the embodiments of the present application can be implemented in software or in hardware. Among them, the name of the module does not constitute a limitation on the unit itself in some cases.

[0159] The readable storage medium provided by the present application is a computer-readable storage medium, and the computer-readable storage medium stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned regression modeling method based on the multi-binning scheme, which can solve the technical problem that the current binning scheme for variables affects the performance and accuracy of the financial risk control model. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as those of the regression modeling method based on the multi-binning scheme provided by the above embodiments, and will not be elaborated here.

[0160] The present application also provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the above-mentioned regression modeling method based on the multi-binning scheme.

[0161] The computer program product provided by the present application can solve the technical problem that the current binning scheme for variables affects the performance and accuracy of the financial risk control model. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as those of the regression modeling method based on the multi-binning scheme provided by the above embodiments, and will not be elaborated here.

[0162] The above are only some embodiments of the present application, and do not limit the patent scope of the present application. All equivalent structural transformations made under the technical concept of the present application by using the content of the specification and drawings of the present application, or directly / indirectly applied to other related technical fields, are included in the patent protection scope of the present application.

Claims

1. A regression modeling method based on a multi-bin scheme, characterized in that, The regression modeling method based on the multi-bin scheme includes: Obtain each original feature variable in the financial customer data, and determine at least two binning schemes corresponding to each of the original feature variables and the WOE coding rules of each of the binning schemes; Convert each of the original feature variables based on each of the WOE coding rules to obtain multiple WOE variables corresponding to each of the original feature variables; Screen the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variables corresponding to each of the original feature variables; Perform regression modeling based on the target WOE variables corresponding to each of the original feature variables to obtain a target risk control model, and the target binning scheme and weight information corresponding to each original feature variable in the target risk control model.

2. The regression modeling method based on a multi-bin scheme according to claim 1, wherein The step of determining at least two binning schemes corresponding to each of the original feature variables and the WOE coding rules of each of the binning schemes includes: Generate an equal-frequency binning scheme and a decision tree binning scheme corresponding to each of the original feature variables, where the equal-frequency binning scheme includes sub-schemes with multiple different quantile values; Obtain the WOE coding rules corresponding to each of the equal-frequency binning schemes and decision tree binning schemes.

3. The regression modeling method based on the multi-bin scheme according to claim 1, wherein After the step of converting each of the original feature variables based on each of the WOE coding rules to obtain multiple WOE variables corresponding to each of the original feature variables, the method further includes: Name each of the WOE variables based on the name of each original feature vector and the attributes of each of the binning schemes, where the attributes include type and / or quantile value; Construct a multi-dimensional binned WOE variable pool based on the named WOE variables, where the multi-dimensional binned WOE variable pool is used for logistic regression modeling and variable screening.

4. The regression modeling method based on the multi-binning scheme according to claim 1, characterized in that, The step of screening the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variables corresponding to each of the original feature variables includes: Initialize an empty initial regression model; Calculate the significance of each WOE variable, and sequentially add the WOE variables with significance higher than the preset addition significance threshold to the initial regression model; and / or, Calculate the significance of the WOE variables already added to the initial regression model, and sequentially remove the WOE variables with significance lower than the preset removal significance threshold; When there is no WOE variable with significance higher than the addition significance threshold among the WOE variables not added to the model, and there is no WOE variable with significance lower than the removal significance threshold among the WOE variables in the model, determine the WOE variables in the initial regression model as the target WOE variables, where the target WOE variables correspond one-to-one with the original feature variables.

5. The regression modeling method based on a multi-bin scheme according to claim 1, characterized in that, The step of screening the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variables corresponding to each of the original feature variables includes: Initialize a LASSO model and the objective function corresponding to the LASSO model, where the objective function includes a regularization parameter and the regression coefficients corresponding to each WOE variable; Train the LASSO model to determine the values of the regression coefficients corresponding to each WOE variable. Determine the target WOE variable with the largest regression coefficient among the WOE variables corresponding to each original feature variable, where the target WOE variable corresponds one-to-one with the original feature variable.

6. The regression modeling method based on a multi-bin scheme according to any one of claims 1 to 5, characterized in that The steps of performing regression modeling according to the target WOE variables corresponding to each original feature variable to obtain a target risk control model, and the target binning scheme and weight information corresponding to each original feature variable in the target risk control model include: Perform logistic regression modeling according to each target WOE variable to obtain a first risk control model. Evaluate the first risk control model through preset verification sample data to obtain the AUC value and KS value corresponding to the first risk control model. If one of the AUC value and KS value does not meet the corresponding preset threshold, return to execute the step: screen the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variables corresponding to each original feature variable. If both the AUC value and KS value meet the corresponding preset thresholds, determine the first risk control model as the target risk control model. Based on the names of the target WOE variables corresponding to each original feature variable, determine the target binning scheme corresponding to each original feature variable. Obtain the weight information corresponding to each original feature variable in the target risk control model.

7. A regression modeling device based on a multi-binning scheme, characterized in that It is characterized in that The regression modeling device based on the multi-binning scheme includes: A scheme determination module, configured to obtain each original feature variable in the financial customer data, and determine at least two binning schemes corresponding to each original feature variable and the WOE coding rules of each binning scheme. A variable coding module, configured to convert each original feature variable based on each WOE coding rule to obtain multiple WOE variables corresponding to each original feature variable. A variable screening module, configured to screen the multiple WOE variables corresponding to each original feature variable to obtain the target WOE variables corresponding to each original feature variable. A model establishment module, configured to perform regression modeling according to the target WOE variables corresponding to each original feature variable to obtain a target risk control model, and the target binning scheme and weight information corresponding to each original feature variable in the target risk control model.

8. An electronic device, characterized in that, The device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the regression modeling method based on the multi-binning scheme according to any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium, and when the computer program is executed by the processor, it implements the steps of the regression modeling method based on the multi-binning scheme according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method according to any one of claims 1 to 6.

Images