A credit approval method and device based on financial credit data

Abstract

The application discloses a credit approval method and device based on financial credit data, adopts various technologies such as data binning, feature screening and logistic regression to perform data screening and modeling, obtains the credit level of a user based on a credit risk model, and matches a corresponding credit approval mode to perform differential data processing; thus, compared with the traditional technology, the application realizes the automatic deployment of credit approval through an algorithm, reduces the labor cost, and the automatic processing scheme enables the application to support real-time and automatic scheduling of differential technical processing flows according to the risk assessment result, thereby improving the system response speed, balancing the resource utilization of the system, and further improving the approval efficiency; therefore, the application is very suitable for large-scale application and popularization.

Description

Technical Field

[0001] This invention belongs to the field of automated data processing technology in financial credit, specifically relating to a credit approval method and apparatus based on financial credit data. Background Technology

[0002] In the fintech field, especially in credit approval scenarios, the volume of user financial credit data processing is rapidly increasing due to the surge in business volume and the increasing complexity of data dimensions. Currently, existing approval methods rely on manual screening and evaluation of financial credit data, resulting in indiscriminate approval. This not only consumes a significant amount of time and manpower but also makes it difficult to automate credit approval processes. Consequently, it is impossible to dynamically and automatically schedule differentiated technical processing flows based on risk assessment results, leading to slow system response, uneven resource utilization, and processing bottlenecks during peak business periods, thus reducing approval efficiency. Therefore, given these shortcomings, providing a credit approval method based on financial credit data that is low in labor costs and highly efficient has become an urgent problem to solve. Summary of the Invention

[0003] The purpose of this invention is to provide a credit approval method and apparatus based on financial credit data to solve the problems of high time and manpower consumption and low efficiency in the existing technology.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a credit approval method based on financial credit data is provided, including: The financial credit data of sample users is obtained and binned to obtain binned data corresponding to different feature indicators in the financial credit data, as well as the WOE value and IV value corresponding to each feature indicator. Based on the IV values ​​of each feature index, feature filtering is performed on each feature index to obtain several optimal feature indices. Using several optimal feature indicators, a logistic regression credit risk prediction model is established, and based on the logistic regression credit risk prediction model, the regression coefficients corresponding to each optimal feature indicator are determined. A credit risk model is constructed based on the regression coefficients corresponding to each optimal feature indicator and the WOE values ​​of each bin data. Obtain the actual financial credit data of the target user, and determine the credit rating of the target user based on the actual financial credit data and the credit risk model; Based on the credit level, the credit approval method for the target user is matched, and the credit approval method is sent to the business end.

[0005] Based on the aforementioned disclosure, after obtaining the financial credit data of sample users, this invention performs binning to obtain binned data corresponding to different feature indicators, as well as the WOE value and IV value corresponding to each binned data and each feature indicator. Then, based on the IV values ​​of each feature indicator, feature filtering is performed to obtain several optimal feature indicators. Next, using the aforementioned optimal feature indicators, a logistic regression credit risk prediction model is established, and the regression coefficients corresponding to each optimal feature indicator are derived from this model. Then, based on the regression coefficients of each optimal feature indicator and the WOE values ​​of each binned data corresponding to each optimal feature indicator, a credit risk model is constructed. Thus, after the model is constructed, it can be used to conduct credit risk assessment, i.e., to obtain the credit rating of the target user based on the target user's actual financial credit data and the credit risk model. Finally, based on the target user's credit rating, a corresponding credit approval method is matched and sent to the business end. Based on this, different approval processes can be triggered within the business end according to the matched credit approval method, thereby achieving automatic and differentiated credit approval processing.

[0006] Through the above design, this invention employs multiple techniques such as data binning, feature filtering, and logistic regression to filter and model data. Based on the obtained credit risk model, it determines the user's credit rating, thereby matching the corresponding credit approval method for differentiated data processing. Thus, compared to traditional technologies, this invention automates credit approval through algorithms, reducing labor costs. Furthermore, this automated processing scheme allows for real-time, automated scheduling of differentiated technical processing flows based on risk assessment results, thereby improving system response speed, balancing system resource utilization, and ultimately increasing approval efficiency. Therefore, this invention is highly suitable for large-scale application and promotion.

[0007] In a possible design, based on the IV values ​​of each feature index, feature filtering is performed on each feature index to obtain several optimal feature indices, including: Multicollinearity checks were performed on each feature indicator to select multiple pre-selected indicators. Remove pre-selected indicators with IV values ​​less than a preset threshold from a number of pre-selected indicators to obtain a number of candidate indicators; A stepwise regression approach is used to screen the candidate indicators to obtain several optimal feature indicators.

[0008] In one possible design, multicollinearity checks are performed on various feature metrics to filter out multiple pre-selected metrics, including: By combining each feature index in pairs, several feature index groups are obtained. Calculate the correlation coefficient between two feature indicators in each feature indicator group; From several groups of feature indicators, select the feature indicator groups whose correlation coefficient is greater than or equal to the correlation threshold, and retain the feature indicator groups whose correlation coefficient is less than the correlation threshold. Each selected feature index group undergoes index elimination processing to obtain several initial feature indices. Using the retained feature index group and several initial feature indices, multiple pre-selected indices were determined.

[0009] In one possible design, each selected feature index group undergoes an index elimination process to obtain several initial feature indices, including: For any selected feature index group, extract the feature index with the largest IV value from the selected feature index group as an initial feature index, and after all selected feature index groups have been queried, several initial feature indices are obtained.

[0010] In one possible design, a stepwise regression approach is used to perform feature filtering on the candidate indicators to obtain several optimal feature indicators, including: Obtain the regression model and the feature set of the non-entered model at the k-th step of regression. When k is 1, the regression model at the k-th step of regression is the initial model, which does not contain any feature variables. The feature set of the non-entered model at the k-th step of regression includes multiple feature variables, and each feature variable corresponds to each candidate indicator. The feature variables that have the greatest impact on the performance of the regression model are selected from the set of features not included in the model, and the selected feature variables are added to the regression model to obtain a new regression model; Based on the significance of the features, the feature variables in the new regression model are subjected to the subsequent term removal process to obtain the updated regression model, and the updated regression model is used as the regression model for the (k+1)th step of regression. Determine whether the iteration termination condition is met; If not, then delete each feature variable in the new regression model from the set of features not included in the model to obtain a new feature set, and use the new feature set as the set of features not included in the model for the (k+1)th step of regression; Increment k by 1 and re-obtain the regression model and the feature set not included in the model at step k until the iteration termination condition is met. Then, take the candidate index corresponding to each feature variable in the regression model that meets the iteration termination condition as the optimal feature index.

[0011] In one possible design, a credit risk model is constructed based on the regression coefficients corresponding to each optimal feature indicator and the WOE values ​​of each bin's data, including: Obtain the rating conversion ratio factor; Based on the score conversion ratio factor, the regression coefficients corresponding to each optimal feature index and the WOE values ​​corresponding to each bin data, the regression scores of each optimal feature index relative to its respective bin data are calculated. The credit risk model is constructed by using the regression scores of each optimal feature index.

[0012] In one possible design, based on the score transformation scaling factor, the regression coefficients corresponding to each optimal feature index, and the WOE values ​​corresponding to each bin, the regression scores of each optimal feature index relative to its respective bin data are calculated, including: The regression scores for each optimal feature index are calculated using the following formula. Score_k = -B×β_k×WOE_k, k=1,2,...,n; In the formula, Score_k represents the regression score of the kth optimal feature indicator, β_k represents the regression coefficient corresponding to the kth optimal feature indicator, WOE_k represents the WOE value of each bin corresponding to the kth optimal feature indicator, n represents the total number of optimal feature indicators, and B represents the score conversion scaling factor, B=PDO / ln(2), and PDO represents a constant. The credit risk model is constructed by using regression scores from each superior bin of data, and includes: The credit risk model is calculated according to the following formula; Total Score =AB×β_0 +ΣScore_k; In the formula, Total Score represents the credit risk model, and represents the credit risk score, A represents the base score, and β_0 represents the intercept term of the logistic regression credit risk prediction model.

[0013] Secondly, a credit approval device based on financial credit data is provided, including: The data binning module is used to acquire financial credit data of sample users and bin the financial credit data to obtain binned data corresponding to different feature indicators in the financial credit data, as well as the WOE value and IV value corresponding to each feature indicator. The feature filtering module is used to perform feature filtering on each feature index based on the IV value of each feature index to obtain several optimal feature indices. The data processing engine is used to establish a logistic regression credit risk prediction model using several optimal feature indicators, and to determine the regression coefficients corresponding to each optimal feature indicator based on the logistic regression credit risk prediction model. The data processing engine is used to construct a credit risk model based on the regression coefficients corresponding to each optimal feature indicator and the WOE value of each bin data. The data processing engine is also used to acquire the actual financial credit data of the target user, and determine the credit rating of the target user based on the actual financial credit data and the credit risk model; The approval routing module is used to match the credit approval method of the target user based on the credit level and send the credit approval method to the business end.

[0014] Thirdly, another credit approval device based on financial credit data is provided. Taking the device as an electronic device as an example, it includes a memory, a processor, and a transceiver that are connected in sequence. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the credit approval method based on financial credit data as described in the first aspect or any possible design in the first aspect.

[0015] Fourthly, a storage medium is provided, on which instructions are stored, which, when executed on a computer, perform the credit approval method based on financial credit data as described in the first aspect or any possible design of the first aspect.

[0016] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer, cause the computer to perform the credit approval method based on financial credit data as described in the first aspect or any possible design of the first aspect.

[0017] Beneficial effects: (1) This invention employs multiple technologies such as data binning, feature filtering, and logistic regression to filter and model data, and uses the obtained credit risk model to determine the user's credit level, thereby matching the corresponding credit approval method for differentiated data processing. Thus, compared with traditional technologies, this invention achieves automated deployment of credit approval through algorithms, which not only reduces labor costs, but also enables this automated processing scheme to support real-time and automated scheduling of differentiated technical processing flows based on risk assessment results, thereby improving system response speed, balancing system resource utilization, and improving approval efficiency. Therefore, this invention is very suitable for large-scale application and promotion. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the steps of a credit approval method based on financial credit data provided in an embodiment of the present invention. Figure 2 A schematic diagram of the structure of a credit approval device based on financial credit data provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0020] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0021] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0022] Example: See Figure 1As shown, the credit approval method based on financial credit data provided in this embodiment employs multiple techniques such as data binning, feature filtering, and logistic regression to filter and model data, thereby determining the user's credit rating and matching the corresponding credit approval method for differentiated data processing. Compared to traditional technologies, this method automates credit approval through algorithms, reducing labor costs. Furthermore, this automation allows the method to support real-time, automated scheduling of differentiated technical processing flows based on risk assessment results, improving system response speed, balancing system resource utilization, and ultimately increasing approval efficiency. Therefore, this method is highly suitable for large-scale application and promotion. For example, this method can, but is not limited to, running on the server and business side. It is understood that the aforementioned execution entity does not constitute a limitation on the embodiments of this application; correspondingly, the operation steps of this method can, but are not limited to, the steps S1 to S6 below.

[0023] S1. Obtain financial credit data of sample users and bin the financial credit data to obtain binned data corresponding to different feature indicators in the financial credit data, as well as the WOE value and IV value corresponding to each feature indicator. In this embodiment, financial credit data of multiple sample users can be obtained, and the financial credit data of any sample user can include, but is not limited to, data corresponding to several feature indicators, such as the gender, age, education level, number of historical loans, outstanding loan balance, number of historical overdue payments, registered capital, years of operation, value-added tax payable, income tax payable, etc. Therefore, the aforementioned gender, age, education level, etc. are all considered as feature indicators.

[0024] After obtaining the aforementioned financial and credit data, data binning can be performed. For example, but not limited to, a decision tree binning algorithm can be used to bin the discrete data in the aforementioned financial and credit data, thereby obtaining binned data corresponding to each different feature indicator, as well as the WOE value and IV value corresponding to each binned data and each feature indicator. Specifically, the binned data of a feature indicator is obtained by dividing the data corresponding to that feature indicator. For example, binning the feature indicator of age can yield binned data for 20 to 30 years old, binned data for 30 to 40 years old, etc. The decision tree binning algorithm is a commonly used data binning processing technique, and its principle will not be elaborated in this embodiment.

[0025] Furthermore, the WOE value for binned data stands for Weight of Evidence, which is an encoding form of the original independent variables. Its calculation formula is as follows: ; in: This refers to the WOE value of the i-th bin of a certain feature indicator. This refers to the number of overdue items in the i-th bin of a certain feature indicator (the number of sample users with overdue items), while Total_bad refers to the total number of overdue items in the sample (i.e., the number of overdue users among all sample users). This refers to the number of non-overdue items in the i-th bin corresponding to a certain feature indicator, while Total_good refers to the total number of non-overdue items in the sample.

[0026] From the formula, WOE is essentially the impact of a certain value of the independent variable on the default rate. Its functions are twofold: First, WOE values ​​are meaningful "standardized" data. As the formula shows, it is a default rate; the higher the default rate of a bin, the higher its WOE value, thus ensuring comparability. This also indicates that the coefficients should be positive during model training. Second, some indicators have a non-linear relationship with the target variable, but after conversion to WOE values, the relationship becomes linear. For linear models like logistic regression, WOE conversion is beneficial for improving model performance.

[0027] Similarly, the IV (information value) measures the predictive power of an independent variable (i.e., a feature indicator). The higher the IV, the stronger the predictive power of the variable, and the more likely it is to be selected by the model. Therefore, the IV can be used to rank variables based on their predictive power, thus enabling variable selection. The formula for calculating the IV is:

[0028] As can be seen from the above formula, the IV value mainly measures the predictive ability by the difference between the default rate and the micro-default rate, and K is the total number of bins corresponding to a certain feature indicator (i.e., the total number of bins).

[0029] Thus, after binning the financial credit data corresponding to each sample user, feature selection and logistic regression modeling are performed, as shown in steps S2 and S3 below.

[0030] S2. Based on the IV values ​​of each feature index, feature screening is performed on each feature index to obtain several optimal feature indices. In specific applications, this embodiment first removes redundant features through multicollinearity check and IV screening. Then, stepwise regression is used to select the most important feature subset until the model performance is optimal, thereby determining the optimal feature index from multiple feature indices. The aforementioned feature screening process can be, but is not limited to, the steps S21 to S23 below.

[0031] S21. Perform multicollinearity checks on each feature indicator to select multiple pre-selected indicators. In practice, the principle of multicollinearity checks is as follows: use the correlation coefficient between variables to identify highly correlated variables (i.e., highly correlated feature indicators). When two or more variables are highly correlated, retain the variable with more information or clearer business meaning, and remove the remaining variables.

[0032] Optionally, the aforementioned multicollinearity check process may be, but is not limited to, as shown in steps S21a to S21e below.

[0033] S21a. Combine each feature index in pairs to obtain several feature index groups; in this embodiment, it is possible, but not limited to, to combine any feature index with the other feature indices, and then poll all feature indices to complete the pairwise combination of each feature index; in this way, after the feature indices are combined in pairs, the correlation coefficient can be calculated, and the process is shown in step S21b below.

[0034] S21b. Calculate the correlation coefficient between two feature indicators in each feature indicator group; in this embodiment, for feature indicator X and feature indicator Y in any feature indicator group, the formula for calculating the correlation coefficient between them is: ; In the formula, This represents the correlation coefficient between feature index X and feature index Y. Let X be the covariance between characteristic index X and characteristic index Y. Let X be the variance. Let Y be the variance.

[0035] Thus, based on the aforementioned formula, the correlation coefficient between the two bins of data in each feature index group can be calculated; then, the bins of data can be filtered according to the correlation coefficient, as shown in step S21c below.

[0036] S21c. From several feature indicator groups, select feature indicator groups with correlation coefficients greater than or equal to the correlation threshold, and retain feature indicator groups with correlation coefficients less than the correlation threshold. In this embodiment, if the correlation coefficient of any feature indicator group is greater than or equal to the correlation threshold (e.g., greater than or equal to 0.8), it indicates that the two feature indicators in that feature indicator group are highly correlated. In this case, the feature indicator with the larger information value needs to be retained. Conversely, if the correlation coefficient is less than the correlation threshold, it can be directly retained. The process of data removal for the selected feature indicator groups is shown in step S21d below.

[0037] S21d. Perform indicator elimination processing on each selected feature indicator group to obtain several initial feature indicators; in this embodiment, for any selected feature indicator group, extract the feature indicator with the largest IV value from any feature indicator group as an initial feature indicator, and after all selected feature indicator groups have been queried, several initial feature indicators are obtained; in this way, after all selected feature indicator groups have been queried in the aforementioned manner, several initial feature indicators can be obtained.

[0038] After completing the data screening of the feature index group with a correlation coefficient greater than or equal to the correlation threshold, several candidate indicators can be determined by combining the aforementioned retained feature index group, as shown in step S21e below.

[0039] S21e. Using the retained feature index group and several initial feature indices, a number of pre-selected indices are determined; in this embodiment, for example, but not limited to, each feature index in the retained feature index group and several initial feature indices can be used as pre-selected indices.

[0040] After completing the multicollinearity check of each characteristic index through the aforementioned steps S21a to S21d, IV value screening can be performed, as shown in step S22 below.

[0041] S22. Delete the pre-selected indicators whose IV value is less than the preset threshold from multiple pre-selected indicators to obtain several candidate indicators; in this embodiment, for feature indicators with an IV value that is too low (<0.02), since their ability to distinguish between defaulting customers and normal customers is weak, they are directly eliminated; thus, after obtaining several candidate indicators, a stepwise regression method can be used to screen the final features, and the process is shown in step S23 below.

[0042] S23. Use stepwise regression to screen the candidate indicators to obtain several optimal indicator values. In specific implementation, for example, but not limited to, the following steps S23a to S23f can be used to screen the optimal indicator values.

[0043] S23a. Obtain the regression model and the feature set of the non-entered model at the k-th step of regression, wherein when k is 1, the regression model at the k-th step of regression is the initial model, which does not contain any feature variables, and the feature set of the non-entered model at the k-th step of regression includes multiple feature variables, and each feature variable corresponds one-to-one with each candidate indicator.

[0044] In this embodiment, when performing the first step of regression, the system is in an initial state. At this time, variables (i.e., candidate indicators) are added step by step, starting from an initial model (which can be an empty model). Similarly, in the initial state, the feature set not included in the model contains all candidate indicators. Based on this, after obtaining the regression model and the feature set not included in the model at the k-th step of regression, the forward selection of feature variables can be performed, as shown in step S23b below.

[0045] S23b. Select the feature variables that have the greatest impact on the performance of the regression model from the set of features not included in the model, and add the selected feature variables to the regression model to obtain a new regression model; in this embodiment, for each feature variable not in the model, calculate the performance index of the regression model after adding it to the logistic regression model, select the variable that makes the model improve the most (i.e., the feature code whose performance index value increases the most) as the feature variable that has the greatest impact on the performance of the regression model, and add it to the model for logistic regression; wherein, for example, the performance index can be, but is not limited to, the KS value and the AUC value.

[0046] Among them, the KS value (Kolmogorov-Smirnov) represents the maximum difference between the cumulative distributions of good and bad customers, and is calculated as: KS=max(TPR - FPR).

[0047] In the formula, TPR (True Positive Rate) represents the cumulative bad user ratio (i.e., the proportion of defaulting users, which is obtained by selecting feature vectors for logistic regression to obtain the default probability, and then determining whether a default has occurred based on the default probability), and FPR represents the cumulative good user ratio (i.e., the proportion of non-defaulting users).

[0048] Furthermore, the AUC value (Area Under the Curve) represents the area under the ROC curve, and the closer the value is to 1, the better the model performance.

[0049] In this way, feature vectors can be selected from the feature set that has not been entered into the model to perform logistic regression, and the KS value and AUC value can be used to evaluate the model performance, thereby obtaining the feature variables that maximize the improvement of model performance.

[0050] Thus, after completing the forward selection of feature vectors based on the aforementioned step S23b, the subsequent elimination of feature vectors can be performed, as shown in step S23c below.

[0051] S23c. Based on feature significance, the feature variables in the new regression model are subjected to posterior term elimination processing to obtain the updated regression model, and the updated regression model is used as the regression model for the (k+1)th step regression. In this embodiment, the significance (p-value) of all currently included feature variables in the model is recalculated, and features that are no longer significant are eliminated, that is, features with p-values ​​greater than the threshold are eliminated. In this way, the posterior term elimination processing of feature variables in the new regression model can be completed. Of course, the p-value is a common method for calculating feature significance, and its calculation process will not be described in detail.

[0052] After the eigenvectors are eliminated, the iteration termination condition can be determined, as shown in step S23d below.

[0053] S23d. Determine whether the iteration termination condition is met; in this embodiment, the iteration termination condition is exemplified as follows: when the model performance index reaches the optimal level, and the change in model performance after adding or removing new feature vectors is less than the magnitude threshold, the iteration stops.

[0054] In this embodiment, if the iteration termination condition is not met, the feature set not included in the model needs to be updated. Then, based on the updated regression model, stepwise regression is performed again until the iteration termination condition is met. The cyclic stepwise regression feature selection process is shown in steps S23e and S23f below.

[0055] S23e. If not, delete each feature variable in the new regression model from the set of features not included in the model to obtain a new feature set, and use the new feature set as the set of features not included in the model for the (k+1)th step of regression; in this embodiment, the selected feature variables are deleted from the set of features not included in the model to obtain a new feature set; then, it can be updated as the set of features not included in the model for the next step of regression; finally, the forward selection and backward elimination are performed again in the aforementioned manner until the aforementioned iteration termination condition is met, then the optimal feature index can be obtained, and the process is shown in step S23f below.

[0056] S23f. Increment k by 1, and re-obtain the regression model and the feature set not included in the model at step k, until the iteration termination condition is met. Then, take the candidate indicators corresponding to each feature variable in the regression model that meets the iteration termination condition as the optimal feature indicators.

[0057] Thus, through the aforementioned steps S23a to S23f, feature selection can be completed to obtain the optimal feature index; then, logistic regression modeling can be performed, the process of which is shown in step S3 below.

[0058] S3. Using several optimal feature indicators, establish a logistic regression credit risk prediction model, and based on the logistic regression credit risk prediction model, determine the regression coefficient corresponding to each optimal feature indicator.

[0059] In this embodiment, logistic regression is a binary classification prediction model widely used in the field of credit risk control, used to estimate the probability of default by sample users, as shown in the following formula: ln(P / (1 - P)) =β_0+β_1×x1+β_2×x2+...+β_k×xk+....+β_n×xn; In the formula, P represents the probability of a sample user defaulting (i.e., the probability of overdue payment), β_0 represents the intercept term of the logistic regression credit risk prediction model, β_k represents the regression coefficient corresponding to the k-th optimal feature indicator, xk represents the k-th optimal feature indicator (i.e., the variable encoded as WOE value), β_n represents the regression coefficient corresponding to the n-th optimal feature indicator, xn represents the n-th optimal feature indicator, and n represents the total number of optimal feature indicators.

[0060] Thus, after performing logistic regression on each optimal feature indicator, the regression coefficients of each optimal feature indicator can be obtained; then, the WOE values ​​of each bin corresponding to each optimal feature indicator can be combined to construct a credit risk model, as shown in step S4 below.

[0061] S4. Construct a credit risk model based on the regression coefficients corresponding to each optimal feature indicator and the WOE values ​​of each bin data. In specific implementation, for example, but not limited to, first obtain the score conversion ratio factor; then, based on the score conversion ratio factor, the regression coefficients corresponding to each optimal feature indicator, and the WOE values ​​corresponding to each bin data, calculate the regression score of each optimal feature indicator relative to its respective bin data; finally, construct the credit risk model using the regression scores of each optimal feature indicator.

[0062] Optionally, for example but not limited to, the following formula can be used to calculate the regression score of each optimal feature index.

[0063] Score_k = -B×β_k×WOE_k, k=1,2,...,n; In the formula, Score_k represents the regression score of the kth optimal feature indicator, β_k represents the regression coefficient corresponding to the kth optimal feature indicator, WOE_k represents the WOE value of each bin corresponding to the kth optimal feature indicator, n represents the total number of optimal feature indicators, and B represents the score conversion scaling factor, B=PDO / ln(2), and PDO represents a constant (with a value of 50).

[0064] In this embodiment, assuming that the kth optimal feature index is age, the WOE values ​​of each bin corresponding to age are substituted into the aforementioned formula to obtain the regression score of age relative to each bin. For example, if the bin data of 20 to 30 years old is substituted into the aforementioned formula, the regression score of age relative to the bin data of 20 to 30 years old can be obtained.

[0065] Thus, after calculating the regression scores of each optimal feature index, a credit risk model can be constructed. For example, a credit risk model can be constructed according to, but is not limited to, the following formula.

[0066] Total Score = A - B×β_0 + Σ Score_k; In the formula, Total Score represents the credit risk model, and represents the credit risk score, A represents the base score (with a value of 500), and β_0 represents the intercept term of the logistic regression credit risk prediction model.

[0067] Thus, based on step S4, a credit risk model can be constructed, and the intervals, WOE values, and scores corresponding to each optimal feature indicator can also be obtained. Based on this, the intervals, WOE values, and scores corresponding to each of the aforementioned optimal feature indicators (i.e., the regression scores of each optimal feature indicator relative to its respective binning data) can be recorded to form a variable mapping table, so that the credit risk level can be classified based on the variable mapping table and the credit risk model. The process is shown in step S5 below.

[0068] S5. Obtain the actual financial credit data of the target user, and determine the credit rating of the target user based on the actual financial credit data and the credit risk model. In this embodiment, the binning data of the target user's corresponding feature indicators is determined based on the actual amount of credit data and the aforementioned variable mapping table, and then the regression scores of each optimal feature indicator corresponding to the target user are obtained. For example, based on the target user's age (e.g., 35 years old), the binning data containing the age range of 35 years old (called the target binning) is determined in the variable mapping table, and then the regression score of age relative to the target binning is found in the table. Of course, the method of finding the regression scores of the target bins corresponding to the other optimal feature indicators is the same, and will not be described in detail here.

[0069] Based on this, after obtaining the regression scores of the bins containing the optimal feature indicators of the target user, these scores can be substituted into the aforementioned credit risk model to obtain the credit risk score of the target user. Finally, the credit level can be determined based on the credit risk score to obtain the credit level of the target user.

[0070] After obtaining the credit rating of the target user, differentiated processing can be carried out, as shown in step S6 below.

[0071] S6. Based on the credit rating, match the credit approval method for the target user and send the credit approval method to the business end; in this embodiment, the aforementioned credit risk score and credit rating can be directly sent to the business end for business personnel to refer to; at the same time, different credit approval methods can also be matched based on the credit rating, the process of which is as follows: Step 1: Based on the credit level, a pre-defined credit level process mapping strategy is invoked to determine the credit approval method corresponding to the target user. Then, based on the real-time load, a computing resource allocation strategy corresponding to the credit approval method is determined. The credit approval method includes a target processing path, and this target processing path is associated with a pre-configured data flow interface. In this embodiment, the credit level process mapping strategy is configured with different credit approval methods corresponding to different credit levels. For example, a high-level user is routed to the credit approval engine corresponding to the first data flow interface for automated approval; a low-level user is routed to the manual review engine corresponding to the second data flow interface for processing. Meanwhile, when the data transfer interface in the credit approval process is the first data transfer interface (i.e., automated processing is used), the remaining computing resources (such as CPU space, free memory, and remaining threads) of the business side can be determined based on the real-time load of the business side. Then, based on the remaining computing resources, the computing resource allocation strategy corresponding to this credit approval is determined (optionally, 70% of the CPU space, free memory, and remaining threads can be allocated to the credit approval engine). Of course, when the data transfer interface in the credit approval process is the second data transfer interface, no computing resource allocation strategy is performed (i.e., it enters the manual review queue).

[0072] Step 2: Based on the target processing path in the credit approval method, the actual financial credit data of the target user is routed to the target asynchronous processing queue in the business terminal, and the credit approval engine or manual review interface module bound to the target asynchronous processing queue is triggered, and a target computing resource allocation strategy is configured for the credit approval service engine; wherein, the target asynchronous processing queue is a message queue corresponding to the data flow interface, and the target computing resource allocation strategy is the computing resource allocation strategy corresponding to the credit approval method; in this way, differentiated approval processing can be performed according to the target processing path and computing resource allocation strategy corresponding to the credit approval method.

[0073] In this way, differentiated technical processing procedures can be scheduled in real time and automatically based on risk assessment results, thereby improving the system's response speed, solving the problem of uneven resource utilization, and thus improving approval efficiency.

[0074] Therefore, through the credit approval method based on financial credit data described in detail in steps S1 to S6 above, this invention employs multiple techniques such as data binning, feature filtering, and logistic regression to perform data filtering and modeling. Based on the obtained credit risk model, it determines the user's credit rating, thereby matching the corresponding credit approval method for differentiated data processing. Thus, compared with traditional technologies, this invention achieves automated deployment of credit approval through algorithms, which not only reduces labor costs, but also enables this automated processing scheme to support real-time and automated scheduling of differentiated technical processing flows based on risk assessment results, thereby improving system response speed, balancing system resource utilization, and ultimately improving approval efficiency. Therefore, this invention is highly suitable for large-scale application and promotion.

[0075] like Figure 2 As shown, the second aspect of this embodiment provides a hardware device for implementing the credit approval method based on financial credit data described in the first aspect of the embodiment, comprising: The data binning module is used to acquire financial credit data of sample users and bin the financial credit data to obtain binned data corresponding to different feature indicators in the financial credit data, as well as the WOE value and IV value corresponding to each feature indicator.

[0076] The feature filtering module is used to perform feature filtering on each feature index based on the IV value of each feature index to obtain several optimal feature indices.

[0077] The data processing engine is used to establish a logistic regression credit risk prediction model using several optimal feature indicators, and to determine the regression coefficients corresponding to each optimal feature indicator based on the logistic regression credit risk prediction model.

[0078] The data processing engine is used to construct a credit risk model based on the regression coefficients corresponding to each optimal feature indicator and the WOE value of each bin data.

[0079] The data processing engine is also used to acquire the actual financial credit data of the target user, and determine the credit rating of the target user based on the actual financial credit data and the credit risk model.

[0080] The approval routing module is used to match the credit approval method of the target user based on the credit level and send the credit approval method to the business end.

[0081] The working process, working details and technical effects of the device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0082] like Figure 3As shown, the third aspect of this embodiment provides another credit approval device based on financial credit data. Taking the device as an electronic device as an example, it includes: a memory, a processor, and a transceiver that are connected in sequence. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the credit approval method based on financial credit data as described in the first aspect of the embodiment.

[0083] For specific examples, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0084] In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. For example, the processor may not be limited to microprocessors of the STM32F105 series, reduced instruction set computer (RISC) microprocessors, x86 architecture processors, or processors with integrated neural network processing units (NPUs). The transceiver may be, but is not limited to, a Wi-Fi transceiver, a Bluetooth transceiver, a General Packet Radio Service (GPRS) transceiver, a ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard) transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. Furthermore, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0085] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0086] The fourth aspect of this embodiment provides a storage medium for storing instructions containing the credit approval method based on financial credit data as described in the first aspect of the embodiment. That is, the storage medium stores instructions that, when executed on a computer, perform the credit approval method based on financial credit data as described in the first aspect of the embodiment.

[0087] The storage medium refers to a carrier for storing data, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0088] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0089] The fifth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the credit approval method based on financial credit data as described in the first aspect of the embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0090] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A credit approval method based on financial credit data, characterized in that, include: The financial credit data of sample users is obtained and binned to obtain binned data corresponding to different feature indicators in the financial credit data, as well as the WOE value and IV value corresponding to each feature indicator. Based on the IV values ​​of each feature index, feature filtering is performed on each feature index to obtain several optimal feature indices. Using several optimal feature indicators, a logistic regression credit risk prediction model is established, and based on the logistic regression credit risk prediction model, the regression coefficients corresponding to each optimal feature indicator are determined. A credit risk model is constructed based on the regression coefficients corresponding to each optimal feature indicator and the WOE values ​​of each bin data. Obtain the actual financial credit data of the target user, and determine the credit rating of the target user based on the actual financial credit data and the credit risk model; Based on the credit level, the credit approval method for the target user is matched, and the credit approval method is sent to the business end.

2. The method according to claim 1, characterized in that, Based on the IV values ​​of each feature index, feature filtering is performed on each feature index to obtain several optimal feature indices, including: Multicollinearity checks were performed on each feature indicator to select multiple pre-selected indicators. Remove pre-selected indicators with IV values ​​less than a preset threshold from a number of pre-selected indicators to obtain a number of candidate indicators; A stepwise regression approach is used to screen the candidate indicators to obtain several optimal feature indicators.

3. The method according to claim 2, characterized in that, Multicollinearity checks were performed on each feature indicator to filter out several pre-selected indicators, including: By combining each feature index in pairs, several feature index groups are obtained. Calculate the correlation coefficient between two feature indicators in each feature indicator group; From several groups of feature indicators, select the feature indicator groups whose correlation coefficient is greater than or equal to the correlation threshold, and retain the feature indicator groups whose correlation coefficient is less than the correlation threshold. Each selected feature index group undergoes index elimination processing to obtain several initial feature indices. Using the retained feature index group and several initial feature indices, multiple pre-selected indices were determined.

4. The method according to claim 3, characterized in that, Each selected feature index group undergoes an index elimination process to obtain several initial feature indices, including: For any selected feature index group, extract the feature index with the largest IV value from the selected feature index group as an initial feature index, and after all selected feature index groups have been queried, several initial feature indices are obtained.

5. The method according to claim 2, characterized in that, A stepwise regression approach is used to perform feature filtering on the candidate indicators to obtain several optimal feature indicators, including: Obtain the regression model and the feature set of the non-entered model at the k-th step of regression. When k is 1, the regression model at the k-th step of regression is the initial model, which does not contain any feature variables. The feature set of the non-entered model at the k-th step of regression includes multiple feature variables, and each feature variable corresponds to each candidate indicator. The feature variables that have the greatest impact on the performance of the regression model are selected from the set of features not included in the model, and the selected feature variables are added to the regression model to obtain a new regression model; Based on the significance of the features, the feature variables in the new regression model are subjected to the subsequent term removal process to obtain the updated regression model, and the updated regression model is used as the regression model for the (k+1)th step of regression. Determine whether the iteration termination condition is met; If not, then delete each feature variable in the new regression model from the set of features not included in the model to obtain a new feature set, and use the new feature set as the set of features not included in the model for the (k+1)th step of regression; Increment k by 1 and re-obtain the regression model and the feature set not included in the model at step k until the iteration termination condition is met. Then, take the candidate index corresponding to each feature variable in the regression model that meets the iteration termination condition as the optimal feature index.

6. The method according to claim 1, characterized in that, Based on the regression coefficients corresponding to each optimal feature indicator and the WOE values ​​of each bin, a credit risk model is constructed, including: Obtain the rating conversion ratio factor; Based on the score conversion ratio factor, the regression coefficients corresponding to each optimal feature index and the WOE values ​​corresponding to each bin data, the regression scores of each optimal feature index relative to its respective bin data are calculated. The credit risk model is constructed by using the regression scores of each optimal feature index.

7. The method according to claim 6, characterized in that, Based on the score conversion scaling factor, the regression coefficients corresponding to each optimal feature index, and the WOE values ​​corresponding to each bin, the regression scores of each optimal feature index relative to its respective bin data are calculated, including: The regression scores for each optimal feature index are calculated using the following formula. Score_k = -B×β_k×WOE_k, k=1,2,...,n; In the formula, Score_k represents the regression score of the kth optimal feature indicator, β_k represents the regression coefficient corresponding to the kth optimal feature indicator, WOE_k represents the WOE value of each bin corresponding to the kth optimal feature indicator, n represents the total number of optimal feature indicators, and B represents the score conversion scaling factor, B=PDO / ln(2), and PDO represents a constant. Accordingly, the credit risk model is constructed by using the regression scores of each excellent score box, which includes: The credit risk model is calculated according to the following formula; Total Score=AB×β_0+ΣScore_k; In the formula, Total Score represents the credit risk model, and represents the credit risk score, A represents the base score, and β_0 represents the intercept term of the logistic regression credit risk prediction model.

8. A credit approval device based on financial credit data, characterized in that, include: The data binning module is used to acquire financial credit data of sample users and bin the financial credit data to obtain binned data corresponding to different feature indicators in the financial credit data, as well as the WOE value and IV value corresponding to each feature indicator. The feature filtering module is used to perform feature filtering on each feature index based on the IV value of each feature index to obtain several optimal feature indices. The data processing engine is used to establish a logistic regression credit risk prediction model using several optimal feature indicators, and to determine the regression coefficients corresponding to each optimal feature indicator based on the logistic regression credit risk prediction model. The data processing engine is used to construct a credit risk model based on the regression coefficients corresponding to each optimal feature indicator and the WOE value of each bin data. The data processing engine is also used to acquire the actual financial credit data of the target user, and determine the credit rating of the target user based on the actual financial credit data and the credit risk model; The approval routing module is used to match the credit approval method of the target user based on the credit level and send the credit approval method to the business end.

9. An electronic device, characterized in that, include: A memory, a processor, and a transceiver are sequentially connected in communication, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the credit approval method based on financial credit data as described in any one of claims 1 to 7.

10. A computer program product containing instructions, characterized in that, When the instructions are executed on the computer, the computer performs the credit approval method based on financial credit data as described in any one of claims 1 to 7.

Images