Interactive box separation technical method based on multiple labels

Abstract

The invention discloses a multi-label-based interactive box separation technical method, which comprises the steps of defining a plurality of risk credit labels according to credit multi-dimensional data of a borrower; based on an automatic box separation model, carrying out automatic box separation operation on the risk credit labels to obtain a multi-dimensional box separation result; calculating a parameter index of each risk credit label based on the multi-dimensional binning result, and comparing and evaluating the parameter indexes to obtain a comparison result; based on the comparison result, optimizing and adjusting box separation through an interactive box separation technology to obtain an optimized box separation result; and performing multi-label modeling according to an optimized binning result. According to the method, the behavior modes of the borrowers at different stages and under overdue conditions are carefully captured, abnormal values and special conditions in data are processed, the accuracy of box separation and the predictive ability of the model are improved, the interpretability and stability of the model are enhanced, and a more accurate and reliable risk management tool is provided for credit risk control.

Description

Technical Field

[0001] The present invention relates to the field of big data credit risk control, and in particular to an interactive binning technology method based on multiple labels. Background Art

[0002] In credit risk control modeling, binning is a common method for processing continuous and categorical variables. It aims to divide the variables into several intervals so that the data in each interval has similar characteristics, which is convenient for model processing and risk assessment. Traditional binning methods mainly include automatic binning and manual binning, and most of them are based on a single label.

[0003] However, the single-label method has limitations in capturing dynamic behavior and risk assessment accuracy. The single-label method usually only considers the behavior of borrowers at a certain point in time or in a specific overdue state, and it is difficult to capture the behavioral changes of borrowers in different MOB (Month on Books) and DPD (Days Past Due) stages, resulting in limited prediction capabilities of the model. In addition, the single-label method cannot effectively distinguish the risk characteristics of new accounts and long-term accounts, affecting the risk identification ability of the model. Although the automated binning technology is highly efficient, it lacks flexibility and business knowledge application, and is difficult to adjust in combination with specific business scenarios. It is impossible to use business knowledge for refined processing, and the processing of outliers and special cases in the data is relatively mechanical, affecting the performance of the model. The automated binning process lacks manual intervention, and the interpretability of the binning results is poor. It is difficult for business personnel to understand and explain the setting logic of each bin, which affects the application effect of the model in actual business. Therefore, traditional single-label and automated binning technologies have many limitations in credit risk control modeling, and it is difficult to meet the high requirements of modern risk control models for accuracy, flexibility and interpretability.

[0004] Therefore, there is an urgent need for an interactive binning technology method based on multiple labels. Summary of the invention

[0005] The present invention provides an interactive binning technology method based on multiple labels to solve the above problems existing in the prior art.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A multi-label based interactive binning method, comprising:

[0008] S101: Define several risk credit labels based on the borrower's multi-dimensional credit data;

[0009] S102: Based on the automated binning model, perform automated binning operations on risk credit labels to obtain multi-dimensional binning results;

[0010] S103: Based on the multi-dimensional binning results, calculate the parameter index of each risk credit label, compare and evaluate the parameter index, and obtain the comparison result;

[0011] S104: Based on the comparison result, the binning is optimized and adjusted by using an interactive binning technology to obtain an optimized binning result;

[0012] S105: Perform multi-label modeling based on the optimized binning results.

[0013] Wherein, step S101 includes:

[0014] S1011: Obtain the borrower's credit multi-dimensional data, which includes credit duration MOB, credit overdue days DPD and credit limit;

[0015] S1012: Based on the preset threshold definition labels, define the combined risk credit labels for the credit multidimensional data to obtain the corresponding risk credit labels, wherein the threshold definition labels include new accounts, medium-term accounts, long-term accounts, not overdue, short-term overdue, medium-term overdue and seriously overdue.

[0016] Wherein, step S102 includes:

[0017] S1021: Select target variables as the analysis object, the target variables include information on whether the borrower defaults, customer grade information, and number of loans;

[0018] S1022: Processing the target variable based on the automated binning model. The automated binning model automatically bins the risk credit labels according to the distribution of the target variable and information gain, and determines the binning interval corresponding to each label;

[0019] S1023: Obtain multi-dimensional binning results, which include binning interval information. The default risks are effectively differentiated in different dimensions through the binning intervals.

[0020] Wherein, step S103 includes:

[0021] S1031: for each automatic binning result under the risk credit label, respectively calculate the parameter indicators of the information value IV, the stability index PSI, and the Gini coefficient GINI, and output the calculation results;

[0022] S1032: After executing the calculation step, all parameter indicators under the current tag are compared, and a combination formed by the comparison results of the parameter indicators is determined;

[0023] S1034: Determine whether the combination is consistent with any one of the plurality of preset combinations. If consistent, the following steps are executed in a loop:

[0024] Randomly determine a new binning result from multiple binning results of the current risk credit label and output it;

[0025] Recalculate the IV, PSI, and GINI parameter indicators for the newly determined binning results and output them;

[0026] Compare all parameter indicators under the current tag and determine the combination formed by the comparison results;

[0027] Return to the step of executing whether the combination is consistent with any one of the plurality of preset combinations;

[0028] If they are inconsistent, a new binning result is randomly determined from the remaining binning results of the current risk credit label and output;

[0029] Return to the step of executing the calculation and comparing the parameter index until the number of times the combination is inconsistent with any one of the plurality of preset combinations reaches a preset number threshold.

[0030] Wherein, step S104 includes:

[0031] S1041: Based on the comparison result, the current binning result is evaluated and an evaluation score is obtained;

[0032] S1042: If the evaluation score reaches a preset threshold, and the key indicators in the indicator comparison result meet the preset standard, the current binning scheme is determined to be the optimal binning scheme based on the comparison result;

[0033] S1043: If the evaluation score does not reach the preset threshold, or the key indicator in the indicator comparison result does not reach the preset standard, the interactive binning module is used to optimize and adjust the current binning:

[0034] Manually adjust the binning boundaries, recalculate related indicators, and obtain optimized binning results;

[0035] Re-comparison and evaluation are performed based on the optimized binning results to determine the scores of the optimized binning results;

[0036] Determine whether the optimized binning result meets the preset standards and thresholds. If so, determine that the binning result is the current optimal binning solution.

[0037] If the optimized binning result still does not meet the preset standard, the next round of interactive binning adjustment will continue according to the preset order of multiple binning schemes until the optimal binning result that meets the standard is obtained.

[0038] Wherein, step S105 includes:

[0039] S1051: Combining the binning results with the credit multi-dimensional data under multiple labels to perform multi-label modeling;

[0040] S1052: Based on the preset label type and the optimized binning result, determine the data features of each label type, and merge these features to form a multidimensional data set;

[0041] S1053: Selecting a corresponding multi-label modeling method according to the multidimensional data set, where the multi-label modeling method includes multi-label logistic regression and multi-label decision tree;

[0042] S1054: When performing multi-label modeling, comprehensively consider the variable characteristics under different labels, including account risk characteristics and user credit behavior patterns, to ensure that the multi-label model comprehensively assesses the borrower's risk;

[0043] S1055: According to the multi-label model, the data of each risk credit label type is traversed one by one, modeled in sequence, and finally a comprehensive risk control model that processes multi-label information is constructed.

[0044] Among them, obtaining the corresponding risk credit label includes:

[0045] Based on the preset credit analysis model, set different threshold definition labels;

[0046] Mapping credit multidimensional data with threshold definition labels, and classifying each user's credit data into corresponding threshold definition labels according to different combinations of MOB, DPD and credit limit;

[0047] Based on the mapped threshold definition labels, risk credit labels are generated according to preset combination rules, where the risk credit labels include new account not overdue, new account short-term overdue, new account medium-term overdue, new account seriously overdue, medium-term account not overdue, medium-term account short-term overdue, medium-term account medium-term overdue, medium-term account seriously overdue, long-term account not overdue, long-term account short-term overdue, long-term account medium-term overdue and long-term account seriously overdue.

[0048] Among them, obtaining multi-dimensional binning results includes:

[0049] Obtain a preset multi-dimensional binning model, the multi-dimensional binning model comprising: multiple first-dimensional binning modules;

[0050] Get the running status of the first dimension binning module, which includes: binning completed and binning incomplete;

[0051] When the running state of the first dimension binning module is binning completion, the first dimension binning module is used as the second dimension binning module;

[0052] Get the completion time of the last binning operation of the second dimension binning module;

[0053] Obtain a binning record library corresponding to the second dimension binning module, and determine the binning record generated after the binning completion time point from the binning record library, where the binning record includes: at least one target variable binned by the second dimension binning module and the corresponding first binning time point;

[0054] Get the second binning time point when the target variable was binned with the third dimension binning module last time;

[0055] Acquire an analysis record library of the target variable, and determine the analysis record generated between the first binning time point and the second binning time point from the analysis record library, wherein the analysis record includes: identifying the type of default risk and the corresponding risk level of at least one target variable;

[0056] Associating the analysis records with the corresponding second dimension binning modules;

[0057] Integrate the analysis records associated with the second dimension binning module, obtain the first binning information item, and associate it with the second dimension binning module;

[0058] The first binning information items associated with each second dimensional binning module are integrated to obtain multi-dimensional binning result information that needs further analysis, thereby completing the acquisition.

[0059] Among them, obtaining the evaluation score includes:

[0060] Based on the comparison results, the preset scoring strategy is applied to comprehensively evaluate the current binning results and calculate the evaluation score of each risk credit label;

[0061] According to the preset scoring algorithm, the assessment scores of each risk credit label are weighted or averaged to obtain the overall assessment score.

[0062] Among them, building a comprehensive risk control model for processing multi-label information includes:

[0063] The data of each risk credit label type is traversed one by one, and independent modeling is performed in turn. The modeling process includes model selection, model training and model verification to build a single-label risk control sub-model that reflects the risk characteristics of each label.

[0064] Based on the constructed single-label risk control sub-model, combined with the correlation between risk credit labels and the global characteristics of user credit behavior, a comprehensive risk control model that processes multi-label information is generated;

[0065] Apply the generated comprehensive risk control model to historical risk credit data for comparison and verification, and adjust and optimize the comprehensive risk control model based on the preset risk assessment standards;

[0066] Based on the final optimized comprehensive risk control model, real-time credit risk assessment is conducted on new or existing users, and corresponding credit decision recommendations are generated. The credit decision recommendations include whether to grant credit and whether to adjust the credit limit.

[0067] Compared with the prior art, the present invention has the following advantages:

[0068] Through multi-label definition, the behavior patterns of borrowers in different MOB (Months on Books) and DPD (Days Past Due) stages can be refined. Compared with the single-label method, it can more accurately reflect the dynamic risk characteristics of borrowers and improve the accuracy of risk assessment. The use of automated binning models and automated binning combined with manual adjustments of interactive binning tools can make full use of data distribution and business knowledge for refined processing. Compared with pure automated binning, interactive binning has higher flexibility and allows real-time adjustments to ensure the optimal binning results. By calculating and comparing IV (Information Value) and PSI (Population Stability The interactive binning module allows users to identify and handle outliers and special cases during the binning process to avoid adverse effects of these data on the overall performance of the model, and to make manual adjustments based on business knowledge to ensure the rationality and effectiveness of the binning results. During the interactive binning process, the basis for setting and adjustment of each bin are recorded to improve the transparency and interpretability of the model, so that the final model not only has good predictive performance, but also can provide a clear basis for business decisions. Through the multi-label method, the risk behavior of borrowers can be more comprehensively evaluated, covering various risk scenarios from short-term to long-term overdue, ensuring that the risk control model has high reliability and accuracy in different actual situations.

[0069] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the present invention.

[0070] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0072] Figure 1It is a flowchart of an interactive binning technology method based on multiple labels in an embodiment of the present invention;

[0073] Figure 2 A flowchart for defining several risk credit labels in an embodiment of the present invention;

[0074] Figure 3 This is a flowchart for obtaining multi-dimensional binning results in an embodiment of the present invention. DETAILED DESCRIPTION

[0075] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0076] The embodiment of the present invention provides a multi-label-based interactive binning technology method, including:

[0077] S101: Define several risk credit labels based on the borrower's multi-dimensional credit data;

[0078] S102: Based on the automated binning model, perform automated binning operations on risk credit labels to obtain multi-dimensional binning results;

[0079] S103: Based on the multi-dimensional binning results, calculate the parameter index of each risk credit label, compare and evaluate the parameter index, and obtain the comparison result;

[0080] S104: Based on the comparison result, the binning is optimized and adjusted by using an interactive binning technology to obtain an optimized binning result;

[0081] S105: Perform multi-label modeling based on the optimized binning results.

[0082] The working principle of the above technical solution is: by analyzing the borrower's multidimensional credit data, a number of risk credit labels are defined. The multidimensional data includes but is not limited to the borrower's credit score, income level, debt situation, repayment history, etc. Through the comprehensive analysis of these data, the borrower's risk level in different dimensions can be identified and classified into different risk labels, such as "new account short-term overdue" and "medium-term account moderate overdue". These labels are used to distinguish borrowers at different stages and overdue situations, so as to more carefully assess their risks. For example, a borrower is defined as "high debt risk" and "low repayment ability" labels.

[0083] Based on the automated binning model, the risk credit labels defined above are binned. Binning is a method of dividing continuous variables into discrete groups. By grouping the dimensional data of risk credit labels, noise can be reduced and the stability and interpretability of the model can be improved. Automated binning automatically calculates the best binning points through algorithms to generate binning results in multiple dimensions. For example, a borrower's credit score can be automatically divided into three bins: "low, medium, and high."

[0084] After obtaining the multi-dimensional binning results, calculate the parameter indicators of each risk credit label, such as the bad debt rate of the bin, the proportion of people in the bin, the information value (IV value), etc. By comparing and evaluating these parameter indicators, the bins with the most distinguishing and predictive power can be identified. For example, comparing the bad debt rates of two different bins, if the bad debt rate of a bin is significantly higher than that of other bins, then the risk identification effect of this bin is better.

[0085] Based on the comparison results of the previous step, the interactive binning technology is used to optimize the binning. Interactive binning is a binning technology that combines manual adjustment with automatic algorithms. Through expert experience and data-driven methods, the binning is fine-tuned to further optimize the risk identification effect. For example, if the boundary of an automatic binning is unreasonable, the binning point can be manually adjusted to make it more consistent with the actual business logic.

[0086] Finally, based on the optimized binning results, multi-label modeling is performed. Multi-label modeling refers to the use of multiple risk credit labels to build a prediction model at the same time, so as to make a more accurate prediction of the borrower's comprehensive risk. Each label corresponds to a different risk dimension, and ultimately a comprehensive risk scoring model is formed.

[0087] The beneficial effects of the above technical solution are as follows: multiple labels are defined according to MOB and DPD, and different risk labels are defined based on the borrower's MOB (Months on Books) and DPD (Days Past Due), such as "short-term overdue for new accounts", "moderate overdue for medium-term accounts", etc. These labels are used to distinguish borrowers at different stages and overdue situations, so as to more carefully assess their risks; for each label, automatic binning is performed, and a supervised binning algorithm (such as decision tree binning) is used to automatically divide the binning interval of the variable according to the target variable (such as default rate) under each label. The automatic binning is performed according to the relationship between the data distribution and the target variable to ensure that the binning is statistically significant; according to the results of the automatic binning, indicators are calculated for each label, and the IV (Information Value) and PSI (Population Value) under each label are calculated. The indicators such as the predictive index and the GINI coefficient are used to evaluate the predictive ability and stability of variables under different labels. IV is used to measure the predictive ability of variables, PSI is used to evaluate the stability of variable distribution, and GINI is used to measure the ability of variables to distinguish between good and bad customers. According to the comparison of indicators of variables on different labels, interactive binning adjustment is used. The interactive binning tool is used to manually adjust the binning according to the indicators under different labels to optimize the binning strategy. Through interactive adjustment, the binning can be refined in combination with business knowledge and actual needs. Compared with automatic binning, interactive binning is more flexible and allows users to make adjustments based on real-time feedback to ensure the optimal binning results. Multi-label modeling is performed based on the final results. The optimized binning results are used in combination with data from different labels to perform multi-label modeling (such as multi-label logistic regression or decision tree). Multi-label modeling can simultaneously consider the risk characteristics of borrowers at different stages and overdue situations to improve the overall predictive ability of the model.

[0088] In another embodiment, step S101 includes:

[0089] S1011: Obtain the borrower's credit multi-dimensional data, which includes credit duration MOB, credit overdue days DPD and credit limit;

[0090] S1012: Based on the preset threshold definition labels, define the combined risk credit labels for the credit multidimensional data to obtain the corresponding risk credit labels, wherein the threshold definition labels include new accounts, medium-term accounts, long-term accounts, not overdue, short-term overdue, medium-term overdue and seriously overdue.

[0091] The working principle of the above technical solution is as follows: First, obtain the multi-dimensional credit data of the borrower, including: MOB (Months on Book), representing the number of months the borrower has held the credit product; DPD (Days Past Due), the number of days of credit overdue; credit limit: the credit limit of the borrower.

[0092] Based on the preset thresholds, the system combines the multi-dimensional credit data to define risk credit labels.

[0093] New account: MOB ≤ 6 months

[0094] Medium-term account: 6 months < MOB ≤ 24 months

[0095] Long-term account: MOB > 24 months

[0096] Not overdue: DPD = 0 days

[0097] Short-term overdue: 0 days < DPD ≤ 30 days

[0098] Medium-term overdue: 30 days < DPD ≤ 90 days

[0099] Seriously overdue: DPD > 90 days

[0100] When the multi-dimensional credit data of a certain borrower is: MOB: 18 months, DPD: 5 days, credit limit: 50,000 yuan;

[0101] For a certain borrower, it will be marked as:

[0102] Medium-term account (18 months)

[0103] Short-term overdue (5 days)

[0104] The target variables selected as the analysis objects are: the information on whether the borrower defaults, the customer level information, and the loan frequency information.

[0105] The beneficial effects of the above technical solution are: collect the MOB and DPD data of the borrower, and define labels according to preset thresholds (such as different time periods of MOB and different day intervals of DPD). For example, MOB 0-6 months is defined as "new account", and DPD 1-30 days is defined as "short-term overdue". Combined, the label "new account short-term overdue" can be defined. In applications under specific scenarios, such as in credit risk control, labels for different overdue stages are defined for newly opened accounts in order to more carefully evaluate the risk behavior of these new accounts. The use of multiple labels can more accurately capture the behavior patterns of borrowers at different stages and under different overdue situations than a single label. A newly opened account may show highly active borrowing behavior in the first few months, while a long-term account may show stable but occasional overdue risks. This distinction can be refined through multiple labels to provide more accurate risk assessment.

[0106] In another embodiment, step S102 includes:

[0107] S1021: Select target variables as the analysis object, the target variables include information on whether the borrower defaults, customer grade information, and number of loans;

[0108] S1022: Processing the target variable based on the automated binning model. The automated binning model automatically bins the risk credit labels according to the distribution of the target variable and information gain, and determines the binning interval corresponding to each label;

[0109] S1023: Obtain multi-dimensional binning results, which include binning interval information. The default risks are effectively differentiated in different dimensions through the binning intervals.

[0110] Among them, determining the binning interval corresponding to each label includes:

[0111] Obtain the data distribution of the target variable, where the target variable is the target value associated with the risky credit label;

[0112] Calculate the distribution characteristics of each distribution interval of the target variable, including the statistical characteristics of frequency distribution, mean, and standard deviation, to determine the distribution characteristics of the target variable;

[0113] Based on the distribution of the target variable, select the corresponding binning method, which includes equal-width binning, equal-frequency binning, information gain-based binning, or adaptive binning.

[0114] According to the distribution of the target variable, the division points between each bin interval are calculated. The division points are selected based on maximizing the information gain. The calculation formula for information gain is:

[0115] IG(X)=H(Y)-H(Y|X)

[0116] Among them, H(Y) is the entropy of the target variable Y, and H(Y|X) is the conditional entropy of the given label X;

[0117] The distribution interval of the target variable is divided based on the information gain to obtain the corresponding bin interval for each label. The bin interval is used to represent the distribution of the target variable in different intervals and its related risks.

[0118] According to the distribution characteristics of the target variable and the information gain results, the risk credit labels are automatically binned to determine the specific binning interval corresponding to each label.

[0119] The information gain-based binning method further includes the following steps:

[0120] Preprocess the target variables, remove outliers, and process missing values ​​to ensure the effectiveness of the binning operation;

[0121] According to the preprocessed target variable, the information gain of each candidate bin point is calculated, and the position with the largest information gain is selected as the split point;

[0122] Based on the calculation result of information gain, the final binning interval is determined, and the binning interval is mapped to the corresponding risk credit label;

[0123] The calculation formula of information gain further includes:

[0124] Calculate the entropy H(Y) of the target variable Y:

[0125]

[0126] Among them, y i represents the i-th value of the target variable, P(y i ) is the target variable value y i The probability of

[0127] Calculate the conditional entropy H(Y|X) of the target variable given label X:

[0128]

[0129] Among them, P(x j ,y i ) is the joint probability of the jth value of label X and the ith value of target variable Y, P(y i |x j ) is given a label X = x j When the target variable Y takes the value of y i The conditional probability of .

[0130] The working principle of the above technical solution is as follows: The system uses an automated binning model to bin the risk credit labels according to the distribution and information gain of the target variable;

[0131] For example, the binning results for the label "medium-term account":

[0132] 6 months < MOB ≤ 12 months

[0133] 12 months < MOB ≤ 18 months

[0134] 18 months < MOB ≤ 24 months

[0135] The binning for "short-term overdue" is as follows:

[0136] 1 day ≤ DPD ≤ 7 days

[0137] 8 days ≤ DPD ≤ 15 days

[0138] 16 days ≤ DPD ≤ 30 days

[0139] Obtain multi-dimensional binning results, including information on each binning interval, which can effectively distinguish default risks in different dimensions.

[0140] Among "medium-term accounts", the accounts with a tenure of 12 - 18 months have the lowest default risk; among "short-term overdue", the accounts with 1 - 7 days of overdue have significantly lower default risk compared to those with 8 - 15 days of overdue.

[0141] The beneficial effects of the above technical solution are as follows: Select the target variable (such as default or not), automatically divide the binning intervals under each label according to the target variable, and perform binning according to the distribution and information gain of the target variable. In the application of specific scenarios, for example, when analyzing the income data of borrowers, use the supervised binning algorithm to automatically divide the income into different intervals to ensure that these intervals are most effective in distinguishing default risks under each label. Improve the efficiency and accuracy of binning, and the preliminary binning results provide a basis for subsequent interactive adjustment.

[0142] In another embodiment, step S103 includes:

[0143] S1031: For the automated binning results under each risk credit label, calculate the parameter indicators of information value IV, stability index PSI, and Gini coefficient GINI respectively, and output the calculation results;

[0144] S1032: After performing the calculation steps, compare all the parameter indicators under the current label, and determine the combination formed by the comparison results of the parameter indicators;

[0145] S1034: Determine whether the combination is consistent with any of the multiple preset combinations. If it is consistent, loop and execute the following steps:

[0146] Randomly determine a new binning result from the multiple binning results of the current risk credit label, and output it;

[0147] Recalculate the parameter indicators of IV, PSI, and GINI for the newly determined binning result, and output them;

[0148] Compare all the parameter indicators under the current label to determine the combination formed by the comparison results;

[0149] Return to execute the step of determining whether the execution judgment combination is consistent with any one of the multiple preset combinations;

[0150] If they are inconsistent, randomly determine a new binning result from the remaining binning results of the current risk credit label, and output it;

[0151] Return to execute the step of calculating and comparing the parameter indicators until the number of times the combination is inconsistent with any one of the multiple preset combinations reaches the preset number threshold.

[0152] The working principle of the above technical solution is: for the automatic binning results of each risk credit label, calculate three important indicators:

[0153] Information Value (IV): An indicator to measure the predictive ability of a variable; Population Stability Index (PSI): An indicator to measure the stability of a variable; Gini Coefficient (GINI): An indicator to measure the discrimination ability of a variable;

[0154] For example, for the binning result of "medium-term account":

[0155] 6 - 12 months: IV = 0.2, PSI = 0.05, GINI = 0.3

[0156] 12 - 18 months: IV = 0.25, PSI = 0.03, GINI = 0.35

[0157] 18 - 24 months: IV = 0.22, PSI = 0.04, GINI = 0.32

[0158] Compare the calculated indicators with the preset standards to form a combined result.

[0159] For example, the preset standards are:

[0160] IV > 0.1: Strong

[0161] 0.02 < PSI < 0.1: Stable

[0162] GINI > 0.3: Good

[0163] Then the combined result of "medium-term account" is: [Strong, Stable, Good]

[0164] The obtained combination result is compared with multiple preset ideal combinations.

[0165] Preset combinations include:

[0166] [Strong, stable, good]

[0167] [Strong, Very Stable, Excellent]

[0168] [Medium, Stable, Good]

[0169] If the current combination is consistent with any preset combination, the loop optimization process will begin.

[0170] If the combination is consistent, the system will:

[0171] a. Randomly select a new binning result

[0172] For example, change "Interim Accounts" to:

[0173] 6-15 months,15-24 months

[0174] b. Recalculate IV, PSI, GINI of the new bin

[0175] c. Compare parameter indicators again to form a new combination

[0176] d. Determine whether the new combination is consistent with the preset combination

[0177] This process is repeated until an inconsistent combination is found.

[0178] If the combination is inconsistent, a new bin is randomly selected from the remaining bin results, the parameter indicators are recalculated and compared, and the process is repeated until the preset number of attempts threshold is reached.

[0179] By calculating and comparing multiple indicators, we ensure that the binning results have good predictive ability, stability and distinguishing ability; randomly selecting new binning results can avoid overfitting the model to a specific binning method; by trying different binning methods multiple times, we try to find a binning result that meets the preset standards but does not fully meet expectations, which represents a better and undiscovered binning method; setting a threshold for the number of attempts to strike a balance between pursuing the optimal solution and computational efficiency.

[0180] The beneficial effects of the above technical solution are as follows: for the automatic binning results under each label, IV, PSI, GINI and other indicators are calculated respectively. IV (Information Value) is used to measure the predictive ability of the variable. The higher the value, the stronger the variable's ability to distinguish the target variable (such as default); PSI (Population Stability Index) is used to evaluate the stability of the variable distribution. A lower PSI value indicates that the variable's distribution between the training set and the test set is stable; the GINI coefficient is used to measure the variable's ability to distinguish between good and bad customers. The higher the value, the stronger the variable's ability to distinguish. In the application of specific scenarios, for example, under the labels of new account short-term overdue and long-term account long-term overdue, the IV, PSI and GINI values ​​of the income variable are calculated respectively, and these indicators are compared to identify the performance of the income variable under different labels. By comparing the indicators under different labels, the performance of the variable at different stages can be identified, providing a basis for subsequent adjustments.

[0181] In another embodiment, step S104 includes:

[0182] S1041: Based on the comparison result, the current binning result is evaluated and an evaluation score is obtained;

[0183] S1042: If the evaluation score reaches a preset threshold, and the key indicators in the indicator comparison result meet the preset standard, the current binning scheme is determined to be the optimal binning scheme based on the comparison result;

[0184] S1043: If the evaluation score does not reach the preset threshold, or the key indicator in the indicator comparison result does not reach the preset standard, the interactive binning module is used to optimize and adjust the current binning:

[0185] Manually adjust the binning boundaries, recalculate related indicators, and obtain optimized binning results;

[0186] Re-comparison and evaluation are performed based on the optimized binning results to determine the scores of the optimized binning results;

[0187] Determine whether the optimized binning result meets the preset standards and thresholds. If so, determine that the binning result is the current optimal binning solution.

[0188] If the optimized binning result still does not meet the preset standard, the next round of interactive binning adjustment will continue according to the preset order of multiple binning schemes until the optimal binning result that meets the standard is obtained.

[0189] The working principle of the above technical solution is: based on the previous comparison results, the current binning results are evaluated and scored.

[0190] For example, evaluation criteria include:

[0191] IV value weight: 40%

[0192] Weight of PSI value: 30%

[0193] Weight of GINI coefficient: 30%

[0194] Assume that the current binning results of "Medium-term Account" are:

[0195] 6-15 months: IV = 0.23, PSI = 0.04, GINI = 0.33

[0196] 15-24 months: IV = 0.25, PSI = 0.03, GINI = 0.35

[0197] Evaluation score calculation:

[0198] Score = 0.4*(0.23+0.25) / 2+0.3*(1-0.04)+0.3*0.35=0.614

[0199] Assuming the preset threshold is 0.6, the key indicator standards are:

[0200] IV>0.2

[0201] PSI<0.05

[0202] GINI>0.3

[0203] In this example, the evaluation score (0.614) reaches the preset threshold (0.6), and all key indicators meet the standards, so the current binning scheme can be determined as the optimal scheme.

[0204] If the evaluation score does not meet the standard or the key indicators do not meet the requirements, interactive optimization is required.

[0205] For example, the initial binning results for "Long-term Accounts" are:

[0206] 24-36 months: IV = 0.18, PSI = 0.06, GINI = 0.28

[0207] 36-48 months: IV = 0.22, PSI = 0.04, GINI = 0.32

[0208] The IV and GINI of this result do not meet the standards and need to be optimized.

[0209] Manually adjust the bin boundaries to: 24-40 months, 40-48 months;

[0210] Recalculate indicators

[0211] The new result is:

[0212] 24-40 months: IV = 0.21, PSI = 0.05, GINI = 0.31

[0213] 40-48 months: IV = 0.23, PSI = 0.04, GINI = 0.33

[0214] Re-evaluation, recalculating scores using the same grading criteria;

[0215] Determine whether it meets the standards. If the new binning result meets all the standards, it can be determined as the optimal solution;

[0216] If the criteria are still not met, other preset binning schemes may be tried, such as: 24-42 months, 42-48 months, and the process is repeated until a binning scheme that meets all criteria is found.

[0217] By giving different weights to different indicators, the quality of binning can be evaluated more accurately. Automatic binning cannot take into account all business factors, and professional judgment and business knowledge can be introduced through manual intervention. Analysts are allowed to adjust binning boundaries according to specific circumstances to adapt to different data characteristics and business needs. Through multiple rounds of adjustment and evaluation, the optimal solution can be gradually approached. Preset standards and thresholds are set to ensure that the final binning solution meets consistent quality requirements. The order of multiple binning solutions is preset to control the time cost of optimization while ensuring quality.

[0218] The beneficial effect of the above technical solution is: using the interactive binning tool, manually adjust the binning boundaries based on the indicators calculated in the previous step. For example, if the IV value of a binning interval under a certain label is significantly lower than that of other intervals, the boundary of the interval can be manually adjusted and the indicators can be recalculated until the optimal binning solution is obtained. In the application of specific scenarios, for example, for credit score variables, the binning intervals are manually adjusted at different MOB stages to ensure that the binning at each stage can maximize the distinction of default risks, and the adjustment process is recorded for explanation and reproduction. Interactive binning allows more refined and flexible adjustments compared to automated binning. Users can make real-time adjustments based on business knowledge and actual needs, handle outliers and special cases in the data, improve the accuracy and rationality of binning, and ultimately improve the predictive ability and interpretability of the model. Through interactive adjustments, binning can be refined based on business knowledge and actual needs. Interactive binning is more flexible than automated binning, allowing users to make adjustments based on real-time feedback to ensure the optimal binning results; at the same time, it improves the flexibility and accuracy of binning, making the model more in line with actual business conditions. Interactive binning combines business knowledge to handle data anomalies and special situations, and can better capture the true risk characteristics of variables.

[0219] In another embodiment, step S105 includes:

[0220] S1051: Combining the binning results with the credit multi-dimensional data under multiple labels to perform multi-label modeling;

[0221] S1052: Based on the preset label type and the optimized binning result, determine the data features of each label type, and merge these features to form a multidimensional data set;

[0222] S1053: Selecting a corresponding multi-label modeling method according to the multidimensional data set, where the multi-label modeling method includes multi-label logistic regression and multi-label decision tree;

[0223] S1054: When performing multi-label modeling, comprehensively consider the variable characteristics under different labels, including account risk characteristics and user credit behavior patterns, to ensure that the multi-label model comprehensively assesses the borrower's risk;

[0224] S1055: According to the multi-label model, the data of each risk credit label type is traversed one by one, modeled in sequence, and finally a comprehensive risk control model that processes multi-label information is constructed.

[0225] The working principle of the above technical solution is as follows: binning is a method of discretizing continuous variables, which is usually used to process features in credit scoring models; this step is to discretize the original continuous data into a form that is easier to understand and process. Multidimensional data means that when evaluating a borrower, there are multiple related data sets, each corresponding to a different label or category; combining the binning results with the multidimensional data under multiple labels is to prepare for the next step of multi-label modeling, ensuring that the data under each label is processed consistently.

[0226] Bin the data under each label so that the features under each label are discretized in a similar way; ensure that the binned results correspond to the data set under each label, forming a data preparation stage for multi-label modeling;

[0227] Determine in advance the different labels or categories that need to be evaluated. These labels are different credit risk levels or other classifications. After binning, optimization processing is required to ensure that the data features of each label type can play the best role in the modeling process. Combine the optimized binning results with the data features of each label type to form a comprehensive multidimensional data set for subsequent modeling.

[0228] According to the preset label type, the binning results are optimized so that the data features of each label type have relatively consistent processing standards; the optimized binning results are merged with the data features of each label type to form a multidimensional data set, preparing for the subsequent multi-label modeling.

[0229] Analyze and evaluate multidimensional data sets and select the most suitable multi-label modeling method; multi-label logistic regression is suitable for situations where labels are independent or have low correlation, while multi-label decision trees can handle situations where there are complex dependencies between labels.

[0230] During the modeling process, it is necessary to comprehensively consider the various variable characteristics under each label type, such as account risk characteristics and user credit behavior patterns; by comprehensively considering different characteristics, it is ensured that the multi-label model can comprehensively assess the borrower's risk, rather than relying solely on a single label or feature.

[0231] Conduct in-depth analysis and mining of variable features under each label type to ensure that key account risk features and user credit behavior patterns are included; during the modeling process, combine the output results of the multi-label model and comprehensively consider risk assessments under all label types to improve overall prediction accuracy and comprehensiveness.

[0232] Model and evaluate each preset risk credit label type in turn; build a comprehensive risk control model that can process and integrate the risk assessment results of each label type.

[0233] The beneficial effects of the above technical solution are: using the final adjusted binning results, merging the data under different labels for multi-label modeling. Selecting appropriate modeling methods (such as multi-label logistic regression, multi-label decision tree), while considering the variable characteristics under different labels. In the application of specific scenarios, for example, in the final model, comprehensively consider the overdue risks of new accounts, medium-term accounts and long-term accounts, and build a comprehensive risk control model that can process multi-label information at the same time, so as to improve the overall prediction ability and stability. Through multi-label modeling, the risk of borrowers can be more comprehensively assessed, and the robustness and stability of the model can be improved.

[0234] In another embodiment, obtaining a corresponding risk credit label includes:

[0235] Based on the preset credit analysis model, set different threshold definition labels;

[0236] Mapping credit multidimensional data with threshold definition labels, and classifying each user's credit data into corresponding threshold definition labels according to different combinations of MOB, DPD and credit limit;

[0237] Based on the mapped threshold definition labels, risk credit labels are generated according to preset combination rules, where the risk credit labels include new account not overdue, new account short-term overdue, new account medium-term overdue, new account seriously overdue, medium-term account not overdue, medium-term account short-term overdue, medium-term account medium-term overdue, medium-term account seriously overdue, long-term account not overdue, long-term account short-term overdue, long-term account medium-term overdue and long-term account seriously overdue.

[0238] The working principle of the above technical solution is: according to the preset combination rules, the mapped threshold definition labels are further combined to form more specific risk credit labels; risk credit labels cover different risk levels, such as new account overdue situation, mid-term account performance, etc.

[0239] According to the preset combination rules, combined with the mapped threshold definition labels, specific risk credit labels are generated, such as new accounts not overdue, short-term overdue, medium-term overdue, etc.

[0240] The credit analysis model and corresponding threshold definitions are set in advance to map the borrower's data to specific risk labels; mapping and classification are performed based on the user's credit data through the set thresholds and rules to ensure that each user can be accurately classified into the appropriate risk label; combined with the preset combination rules, detailed risk credit labels are generated for subsequent risk assessment and modeling.

[0241] The beneficial effects of the above technical solution are: the implementation based on the preset credit analysis model and threshold definition labels not only improves the accuracy and efficiency of credit risk assessment, but also provides financial institutions with better risk management tools and enhances customer relationship management capabilities, thereby promoting the sustainable development of the overall business.

[0242] In another embodiment, obtaining a multi-dimensional binning result includes:

[0243] Obtain a preset multi-dimensional binning model, the multi-dimensional binning model comprising: multiple first-dimensional binning modules;

[0244] Get the running status of the first dimension binning module, which includes: binning completed and binning incomplete;

[0245] When the running state of the first dimension binning module is binning completion, the first dimension binning module is used as the second dimension binning module;

[0246] Get the completion time of the last binning operation of the second dimension binning module;

[0247] Obtain a binning record library corresponding to the second dimension binning module, and determine the binning record generated after the binning completion time point from the binning record library, where the binning record includes: at least one target variable binned by the second dimension binning module and the corresponding first binning time point;

[0248] Get the second binning time point when the target variable was binned with the third dimension binning module last time;

[0249] Acquire an analysis record library of the target variable, and determine the analysis record generated between the first binning time point and the second binning time point from the analysis record library, wherein the analysis record includes: identifying the type of default risk and the corresponding risk level of at least one target variable;

[0250] Associating the analysis records with the corresponding second dimension binning modules;

[0251] Integrate the analysis records associated with the second dimension binning module, obtain the first binning information item, and associate it with the second dimension binning module;

[0252] The first binning information items associated with each second dimensional binning module are integrated to obtain multi-dimensional binning result information that needs further analysis, thereby completing the acquisition.

[0253] The working principle of the above technical solution is: multiple dimensional binning modules are preset, each module is responsible for a specific data analysis task; the first dimension binning module is responsible for the initial data analysis and classification. According to the preset model settings, multiple first dimension binning modules are obtained, and each module analyzes different data dimensions.

[0254] Monitor and obtain the running status of the first-dimension binning module to determine when to proceed to the next step; when the running status of the first-dimension binning module is binning completed, use it as the second-dimension binning module; when the first-dimension binning module completes the task, its result can be used as the input data of the second-dimension binning module; ensure the continuity and integrity of data analysis so that the analysis of each dimension can be interrelated and influence each other.

[0255] If the running status of the first-dimension binning module shows completion, its output data is passed to the second-dimension binning module of the next stage for further processing, and the completion time point of the last binning operation of the second-dimension binning module is obtained; the completion time of each binning operation is recorded to determine the order and interval of data processing; the operation history of each module can be traced back through the time point, which is helpful for analyzing and evaluating the model effect; the exact time when the second-dimension binning module last completed the binning operation is obtained for subsequent analysis; the binning record library corresponding to the second-dimension binning module is obtained; the analysis results and related data of each module are stored for subsequent comparison and further analysis to ensure the security and accuracy of the data, as well as the convenience of subsequent data access; the binning record library of the second-dimension binning module is accessed and queried to retrieve the relevant analysis records generated after a specific time point; the second binning time point when the target variable was last binned with the third-dimension binning module is obtained.

[0256] Record the analysis process and timeline of each target variable to track and analyze risk changes; determine the data generation and processing time of each stage to ensure the continuity and consistency of operations; determine the time point of the last analysis operation of the target variable in order to compare and integrate with the relevant data of the third-dimensional binning module; include detailed analysis results and risk assessments of each target variable. The record library can be used to track and analyze data changes and trends, query the analysis record library of the target variable to obtain relevant analysis records generated between the first binning time point and the second binning time point, and associate the analysis records with the corresponding second-dimensional binning module; match and integrate the data and analysis results generated by different modules for further data processing and decision support, and combine the analysis results of each module to form a comprehensive multi-dimensional analysis report.

[0257] Associate the analysis records of the target variable with the data of the second-dimensional binning module to ensure the integrity and reliability of the analysis results, and integrate the first binning information items associated with each second-dimensional binning module; aggregate the information and data items from different binning modules into a unified data structure to ensure that the integrated data can accurately reflect the results and relevance of multiple analysis stages, integrate the first binning information items associated with each second-dimensional binning module, and form complete multi-dimensional binning result information. After completing the above steps, integrate and prepare the data for final analysis and reporting to ensure that the entire process is completed smoothly and the data is ready for further decision-making and action; complete all steps of acquiring and integrating data to ensure that the generated multi-dimensional binning result information can be used for subsequent in-depth analysis and application.

[0258] The beneficial effects of the above technical solution are: it demonstrates the operation process of a multi-dimensional binning model in practical application, from data acquisition and analysis to final integration and application. This model can help institutions better understand and manage complex data structures, and support accurate decision-making and the formulation of risk management strategies.

[0259] In another embodiment, obtaining the evaluation score includes:

[0260] Based on the comparison results, the preset scoring strategy is applied to comprehensively evaluate the current binning results and calculate the evaluation score of each risk credit label;

[0261] According to the preset scoring algorithm, the assessment scores of each risk credit label are weighted or averaged to obtain the overall assessment score.

[0262] The working principle of the above technical solution is: before conducting a comprehensive evaluation, it is first necessary to obtain the comparison data of the current binning results; according to the preset scoring strategy, the current binning results are evaluated. The scoring strategy usually includes multiple dimensions, such as default risk, credit history, financial status, etc. Each dimension will set different weights according to its importance and influence.

[0263] A comprehensive assessment is conducted on each risk credit label (for example, new account not overdue, new account short-term overdue, etc.) and its assessment score is calculated. This process involves combining the scoring results of various dimensions to form a comprehensive score.

[0264] According to the preset scoring algorithm, the evaluation scores of each risk credit label are weighted or averaged. Weighted processing is to assign different weights according to the importance of each label, while average processing is to simply average the scores of all labels.

[0265] Finally, an overall assessment score is obtained through weighted or averaging processing, which can reflect the comprehensive assessment results of the current credit risk status and provide a basis for subsequent decision-making.

[0266] The beneficial effects of the above technical solution are as follows: the application of preset scoring strategies and algorithms can ensure the standardization and systematization of the evaluation process, thereby improving the accuracy and reliability of the evaluation results; through the comprehensive evaluation of risk credit labels, financial institutions can more clearly identify potential risky customers, take timely measures to reduce credit risks, and optimize risk management strategies; the overall evaluation score provides data support for decisions such as credit approval, customer management and risk control, helping decision makers to make more scientific and reasonable judgments; through comprehensive evaluation, financial institutions can design and provide more personalized credit products based on customers' risk scores to meet the needs of different customers; transparent evaluation processes and data-based decisions can enhance customers' trust in financial institutions, improve customer satisfaction, and promote the establishment of long-term cooperative relationships; through the evaluation of risk credit labels, financial institutions can allocate resources more effectively, pay more attention and support to high-risk customers, thereby improving overall operational efficiency; the evaluation results can be used as a feedback mechanism to help financial institutions continuously optimize scoring strategies and algorithms and enhance future risk assessment capabilities.

[0267] In another embodiment, a comprehensive risk control model for processing multi-label information is constructed, including:

[0268] The data of each risk credit label type is traversed one by one, and independent modeling is performed in turn. The modeling process includes model selection, model training and model verification to build a single-label risk control sub-model that reflects the risk characteristics of each label.

[0269] Based on the constructed single-label risk control sub-model, combined with the correlation between risk credit labels and the global characteristics of user credit behavior, a comprehensive risk control model that processes multi-label information is generated;

[0270] Apply the generated comprehensive risk control model to historical risk credit data for comparison and verification, and adjust and optimize the comprehensive risk control model based on the preset risk assessment standards;

[0271] Based on the final optimized comprehensive risk control model, real-time credit risk assessment is conducted on new or existing users, and corresponding credit decision recommendations are generated. The credit decision recommendations include whether to grant credit and whether to adjust the credit limit.

[0272] The working principle of the above technical solution is: first, for various risk credit labels (such as new accounts not overdue, new accounts short-term overdue, etc.), separate modeling is carried out, which includes selecting appropriate modeling algorithms (such as logistic regression, decision trees, etc.), using historical data for model training, and ensuring the accuracy and reliability of the model through verification; after establishing a single-label risk control sub-model, it is necessary to integrate the various sub-models into a comprehensive risk control model. This step takes into account the correlation between different risk labels and the global characteristics of user credit behavior, such as the user's overall credit history, financial status, etc.

[0273] The generated comprehensive risk control model is applied to historical data for verification to confirm the performance of the model on real data. The model is adjusted and optimized through preset risk assessment standards to ensure that the model can accurately reflect the risk characteristics of each label.

[0274] The final optimized comprehensive risk control model can be used to conduct real-time credit risk assessment for new or existing users. This process generates credit decision recommendations, such as whether to grant credit and recommendations for adjusting credit limits.

[0275] The beneficial effects of the above technical solutions are as follows: the establishment of a single-label risk control sub-model and a comprehensive risk control model enables financial institutions to more accurately identify risky customers under different risk labels, which helps to reduce default rates and improve asset quality; based on model evaluation, financial institutions can provide each customer with personalized credit products and services to meet the specific needs of customers and improve customer satisfaction and loyalty; the application of the comprehensive risk control model makes credit decisions more scientific and efficient, and the automated evaluation process reduces the need for manual intervention, speeds up decision-making, and reduces operating costs; the verification and optimization process of the model is not just a one-time process, but also includes continuous monitoring and feedback of the model. By monitoring the performance of the model in actual applications, the model can be adjusted and optimized in a timely manner to maintain the continued effectiveness of its predictive ability; the establishment and application of the comprehensive risk control model enhances the overall risk management capabilities of financial institutions. By better controlling credit risks, institutions can steadily develop their businesses and enhance their market competitiveness.

[0276] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A multi-label based interactive binning method, characterized in that: include: S101: Define several risk credit labels based on the borrower's multi-dimensional credit data; S102: Based on the automated binning model, perform automated binning operations on risk credit labels to obtain multi-dimensional binning results; S103: Based on the multi-dimensional binning results, calculate the parameter index of each risk credit label, compare and evaluate the parameter index, and obtain the comparison result; S104: Based on the comparison result, the binning is optimized and adjusted by using an interactive binning technology to obtain an optimized binning result; S105: Perform multi-label modeling based on the optimized binning results.

2. According to the method of claim 1, the method is characterized in that: Step S101 includes: S1011: Obtain the borrower's credit multi-dimensional data, which includes credit duration MOB, credit overdue days DPD and credit limit; S1012: Based on the preset threshold definition labels, define the combined risk credit labels for the credit multidimensional data to obtain the corresponding risk credit labels, wherein the threshold definition labels include new accounts, medium-term accounts, long-term accounts, not overdue, short-term overdue, medium-term overdue and seriously overdue.

3. The method of interactive binning based on multiple labels according to claim 1, characterized in that: Step S102 includes: S1021: Select target variables as the analysis object, the target variables include information on whether the borrower defaults, customer grade information, and number of loans; S1022: Processing the target variable based on the automated binning model. The automated binning model automatically bins the risk credit labels according to the distribution of the target variable and information gain, and determines the binning interval corresponding to each label; S1023: Obtain multi-dimensional binning results, which include binning interval information. The default risks are effectively differentiated in different dimensions through the binning intervals.

4. The method of interactive binning based on multiple labels according to claim 1, characterized in that: Step S103 includes: S1031: for each automatic binning result under the risk credit label, respectively calculate the parameter indicators of the information value IV, the stability index PSI, and the Gini coefficient GINI, and output the calculation results; S1032: After executing the calculation step, all parameter indicators under the current tag are compared, and a combination formed by the comparison results of the parameter indicators is determined; S1034: Determine whether the combination is consistent with any one of the plurality of preset combinations. If consistent, the following steps are executed in a loop: Randomly determine a new binning result from multiple binning results of the current risk credit label and output it; Recalculate the IV, PSI, and GINI parameter indicators for the newly determined binning results and output them; Compare all parameter indicators under the current tag and determine the combination formed by the comparison results; Return to the step of executing whether the combination is consistent with any one of the plurality of preset combinations; If they are inconsistent, a new binning result is randomly determined from the remaining binning results of the current risk credit label and output; Return to the step of executing the calculation and comparing the parameter index until the number of times the combination is inconsistent with any one of the plurality of preset combinations reaches a preset number threshold.

5. The method of interactive binning based on multiple labels according to claim 1, characterized in that: Step S104 includes: S1041: Based on the comparison result, the current binning result is evaluated and an evaluation score is obtained; S1042: If the evaluation score reaches a preset threshold, and the key indicators in the indicator comparison result meet the preset standard, the current binning scheme is determined to be the optimal binning scheme based on the comparison result; S1043: If the evaluation score does not reach the preset threshold, or the key indicator in the indicator comparison result does not reach the preset standard, the interactive binning module is used to optimize and adjust the current binning: Manually adjust the binning boundaries, recalculate related indicators, and obtain optimized binning results; Re-comparison and evaluation are performed based on the optimized binning results to determine the scores of the optimized binning results; Determine whether the optimized binning result meets the preset standards and thresholds. If so, determine that the binning result is the current optimal binning solution. If the optimized binning result still does not meet the preset standard, the next round of interactive binning adjustment will continue according to the preset order of multiple binning schemes until the optimal binning result that meets the standard is obtained.

6. The method of interactive binning based on multiple labels according to claim 1, characterized in that: Step S105 includes: S1051: Combining the binning results with the credit multi-dimensional data under multiple labels to perform multi-label modeling; S1052: Based on the preset label type and the optimized binning result, determine the data features of each label type, and merge these features to form a multidimensional data set; S1053: Selecting a corresponding multi-label modeling method according to the multidimensional data set, where the multi-label modeling method includes multi-label logistic regression and multi-label decision tree; S1054: When performing multi-label modeling, comprehensively consider the variable characteristics under different labels, including account risk characteristics and user credit behavior patterns, to ensure that the multi-label model comprehensively assesses the borrower's risk; S1055: According to the multi-label model, the data of each risk credit label type is traversed one by one, modeled in sequence, and finally a comprehensive risk control model that processes multi-label information is constructed.

7. The method of interactive binning based on multiple labels according to claim 2, characterized in that: Obtain the corresponding risk credit label, including: Based on the preset credit analysis model, set different threshold definition labels; Mapping credit multidimensional data with threshold definition labels, and classifying each user's credit data into corresponding threshold definition labels according to different combinations of MOB, DPD and credit limit; Based on the mapped threshold definition labels, risk credit labels are generated according to preset combination rules, where the risk credit labels include new account not overdue, new account short-term overdue, new account medium-term overdue, new account seriously overdue, medium-term account not overdue, medium-term account short-term overdue, medium-term account medium-term overdue, medium-term account seriously overdue, long-term account not overdue, long-term account short-term overdue, long-term account medium-term overdue and long-term account seriously overdue.

8. The method of interactive binning based on multiple labels according to claim 3, characterized in that: Get multi-dimensional binning results, including: Obtain a preset multi-dimensional binning model, the multi-dimensional binning model comprising: multiple first-dimensional binning modules; Get the running status of the first dimension binning module, which includes: binning completed and binning incomplete; When the running state of the first dimension binning module is binning completion, the first dimension binning module is used as the second dimension binning module; Get the completion time of the last binning operation of the second dimension binning module; Obtain a binning record library corresponding to the second dimension binning module, and determine the binning record generated after the binning completion time point from the binning record library, where the binning record includes: at least one target variable binned by the second dimension binning module and the corresponding first binning time point; Get the second binning time point when the target variable was binned with the third dimension binning module last time; Acquire an analysis record library of the target variable, and determine the analysis record generated between the first binning time point and the second binning time point from the analysis record library, wherein the analysis record includes: identifying the type of default risk and the corresponding risk level of at least one target variable; Associating the analysis records with the corresponding second dimension binning modules; Integrate the analysis records associated with the second dimension binning module, obtain the first binning information item, and associate it with the second dimension binning module; The first binning information items associated with each second dimensional binning module are integrated to obtain multi-dimensional binning result information that needs further analysis, thereby completing the acquisition.

9. The method of interactive binning based on multiple labels according to claim 5, characterized in that: Get assessment scores, including: Based on the comparison results, the preset scoring strategy is applied to comprehensively evaluate the current binning results and calculate the evaluation score of each risk credit label; According to the preset scoring algorithm, the assessment scores of each risk credit label are weighted or averaged to obtain the overall assessment score.

10. The method of interactive binning based on multiple labels according to claim 6, characterized in that: Build a comprehensive risk control model that processes multi-label information, including: The data of each risk credit label type is traversed one by one, and independent modeling is performed in turn. The modeling process includes model selection, model training and model verification, so as to construct a single-label risk control sub-model that reflects the risk characteristics of each label; based on the constructed single-label risk control sub-model, combined with the correlation between each risk credit label and the global characteristics of user credit behavior, a comprehensive risk control model that processes multi-label information is integrated and generated; Apply the generated comprehensive risk control model to historical risk credit data for comparison and verification, and adjust and optimize the comprehensive risk control model based on the preset risk assessment standards; Based on the final optimized comprehensive risk control model, real-time credit risk assessment is conducted on new or existing users, and corresponding credit decision recommendations are generated. The credit decision recommendations include whether to grant credit and whether to adjust the credit limit.

Images