Financial overdue risk intelligent identification and analysis method

By performing cluster analysis and dynamic credit scores on lenders' credit data, combined with behavior analysis under economic pressure, and building a predictive model, it solves the problem that traditional credit scoring models are difficult to adjust and capture nonlinear features in real time, and increases the risks of financial institutions Management efficiency and forecast accuracy.

CN118941379AInactive Publication Date: 2024-11-12SHENZHEN HONGZHENG BUSINESS SERVICE CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411073132.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The traditional static credit scoring model is difficult to adjust the lender's credit rating in real time, resulting in the credit score being out of sync with the lender's current credit status, affecting the risk management and decision-making efficiency of financial institutions, and it is difficult to capture the nonlinearity of lenders in different financial scenarios. Characteristics and loss aversion behavior affect the accuracy of risk prediction.

Method used

By clustering analysis of lenders’ credit data, a dynamic credit grading model is generated, the changes in credit activities are tracked, behavior preferences under economic pressure are analyzed, prediction models are constructed, overdue probability is predicted, risk levels are divided, influencing factors are identified, and correlation is carried out. Analysis and optimize risk prediction.

Benefits of technology

It realizes the high adaptability and dynamic nature of the credit score model, promptly reflects changes in market and financial conditions, improves the sensitivity and accuracy of overdue risk prediction, and allows financial institutions to take measures in advance to prevent risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941379A_ABST
    Figure CN118941379A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of financial risk management, in particular to a financial overdue risk intelligent identification and analysis method, which comprises the following steps of: performing clustering analysis on credit data of loaners, identifying loaner groups of various credit grades, determining a clustering number by utilizing an evaluation index, and generating a loaner credit grading model. According to the method, dynamic clustering is carried out through the Gaussian mixture model, and the risk prediction model is optimized in combination with the cumulative foreground theory, so that the credit scoring model has high adaptability and dynamics, changes of market and personal financial conditions can be reflected in time, and it is ensured that the credit rating of a loan accurately reflects the current credit state in real time. Meanwhile, by deeply analyzing behaviors of the loan in different financial scenes, the overdue risk prediction sensitivity and accuracy are improved, and a financial institution is allowed to take corresponding measures before the loan begins to be overdue.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of financial risk management, and in particular to a method for intelligently identifying and analyzing financial overdue risks. Background Art

[0002] The field of financial risk management technology includes methods and tools for identifying, assessing, monitoring and mitigating various risks in financial markets. The field uses data analysis technology, artificial intelligence, machine learning and big data technology to improve the accuracy and efficiency of risk management, enabling financial institutions to better predict and prevent potential risk events.

[0003] Among them, the intelligent identification and analysis method of financial overdue risk is a specific practice in the field of financial risk management technology, focusing on the use of intelligent technology to identify and analyze the overdue risk of financial products (such as loans and credit cards). The purpose is to help financial institutions take preventive measures by identifying potential default or overdue payment risks at an early stage, thereby reducing losses and optimizing asset quality.

[0004] Although existing technologies have made significant progress in dealing with financial risk management, when the borrower's financial situation or market conditions change, traditional static credit scoring models are difficult to adjust the borrower's credit rating in real time, which can easily lead to the credit score being out of sync with the borrower's current real credit situation, affecting the risk management and decision-making efficiency of financial institutions. In addition, when evaluating the behavioral responses of borrowers in different financial scenarios, traditional models are difficult to accurately capture the nonlinear characteristics and loss aversion behaviors in the decision-making process. For example, when a borrower who repays the loan on time accumulates to a certain level of economic pressure, he may suddenly delay or stop repayment. At the same time, the borrower may excessively reduce consumption and investment when facing potential financial difficulties, affecting the application effect and accuracy of the risk prediction model in actual financial decision-making. Summary of the invention

[0005] This application provides an intelligent identification and analysis method for financial overdue risks, which solves the problem that although the existing technology has made significant progress in dealing with financial risk management, when the financial situation or market conditions of the borrower change, the traditional static credit scoring model is difficult to adjust the credit rating of the borrower in real time, which easily leads to the credit score being out of sync with the borrower's current real credit situation, affecting the risk management and decision-making efficiency of financial institutions. In addition, when evaluating the behavioral responses of borrowers in different financial scenarios, traditional models are difficult to accurately capture the nonlinear characteristics and loss aversion behaviors in the decision-making process. For example, when a borrower who repays the loan on time accumulates to a certain level of economic pressure, he may suddenly start to delay or stop repayment. At the same time, the borrower may excessively reduce consumption and investment when facing potential financial difficulties, affecting the application effect and accuracy of the risk prediction model in actual financial decision-making.

[0006] In view of the above problems, the present application provides a method for intelligent identification and analysis of financial overdue risks.

[0007] The present application provides a method for intelligently identifying and analyzing financial overdue risks, wherein the method comprises:

[0008] S1: By clustering the lender credit data, identify lender groups with multiple credit ratings, determine the number of clusters using evaluation indicators, and generate a lender credit rating model;

[0009] S2: Based on the lender credit grading model, track the lender's credit activity data, capture changes in credit activities and record key events, and generate a dynamic deviation record of the lender's credit;

[0010] S3: Analyze the behavior preference data of the lender under economic pressure and the behavior trend that causes overdue payment through the credit dynamic deviation record of the lender, and generate the overdue behavior dynamic analysis result;

[0011] S4: Using the dynamic analysis results of the overdue behavior, a prediction model is constructed to predict the overdue probability of the lender, and the overdue risk analysis results of the lender are obtained;

[0012] S5: According to the overdue risk analysis results of the lender, the lender risk level is divided, and the key factors leading to the change of overdue probability are analyzed to generate risk classification and overdue factor identification results;

[0013] S6: Analyze the correlation between the overdue factors in the risk classification and overdue factor identification results, identify the interaction between each factor and its impact on the overdue risk, and obtain the risk factor correlation analysis results.

[0014] Preferably, the lender credit grading model includes credit score ranges, overdue record classifications and financial health indexes, the lender credit dynamic deviation records include credit score decline events, the number of new loan applications and the frequency of overdue occurrences, the overdue behavior dynamic analysis results include changes in consumption patterns under economic pressure, abnormal expenditure increases and credit card usage frequencies, the lender overdue risk analysis results include the overdue probability value of each lender, the predicted time period for overdue occurrence and the overdue risk level classification, the risk classification and overdue factor identification results include lender groups graded by overdue probability, overdue economic factors and personal behavioral characteristics, the risk factor correlation analysis results include the correlation strength between financial status and personal behavioral characteristics, the interaction pattern between multiple economic factors and the analysis of complex factors affecting overdue risks.

[0015] Preferably, the step of performing cluster analysis on the lender credit data, identifying lender groups of multiple credit ratings, determining the number of clusters using evaluation indicators, and generating a lender credit rating model further includes:

[0016] S101: By clustering the credit data of lenders, using the Gaussian mixture model and the expectation maximization algorithm to monitor repayment behavior and account information, identify credit behavior patterns, determine the number of distributions through the Bayesian information criterion, and generate behavior pattern classification results;

[0017] S102: Based on the behavior pattern classification result, analyzing the repayment frequency and loan amount usage, identifying multiple credit rating features including payment behavior, account age and credit utilization rate, and generating credit rating difference analysis results;

[0018] S103: Using the credit rating difference analysis results, adjusting the classification standards, optimizing the credit rating group division, and generating a lender credit grading model.

[0019] Preferably, the Gaussian mixture model and the expectation maximization algorithm are in accordance with the improved formula I:

[0020]

[0021] Segment and identify financial overdue risks by referring to lender characteristics and credit rating information;

[0022] Among them, Q'(θ|θ (t) ) is given the current parameter estimate θ (t) After that, the improved conditional expected log-likelihood function, w ij represents the weight coefficient of the i-th data point under the j-th Gaussian distribution, γ(z ij ) is the posterior probability that the i-th data point comes from the j-th Gaussian distribution, α, β, δ are coefficients used to adjust the model complexity and the influence of data features, πj is the mixing coefficient of the jth distribution, ∑ j is the covariance matrix of the jth Gaussian distribution, x i is a data point, μ j is the mean of the jth Gaussian distribution, k is the number of distributions, n is the number of data points, L(x i , C) is a data point x i and the additional term of its credit rating C, λ is the additional term coefficient;

[0023] The Bayesian Information Criterion follows Formula II:

[0024] BIC=-2ln(L)+kln(n)

[0025] Determine the optimal distribution quantity and generate behavior pattern classification results;

[0026] Among them, L is the maximum value of the likelihood function of the model, k is the number of parameters in the model, n is the number of data points, ln represents the natural logarithm function, and BIC penalizes the value of the likelihood function of the model.

[0027] Preferably, the step of tracking the lender's credit activity data, capturing changes in credit activity and recording key events based on the lender's credit grading model, and generating a lender's credit dynamic deviation record further includes:

[0028] S201: using the lender credit grading model, monitoring the daily credit account activities of the lender, including current loan applications and repayment behaviors, and generating daily credit activity records;

[0029] S202: Screening abnormal activities from the daily credit activity records, including a decrease in credit score or a new overdue record, and marking them as key credit change events, generating key credit event records;

[0030] S203: Analyze the key credit event records, sort out the changing trends and key deviation points of credit activities, evaluate the health of credit activities, and generate the lender's credit dynamic deviation record.

[0031] Preferably, the step of analyzing the behavior preference data of the lender under economic pressure and the behavior trend causing overdue payment through the credit dynamic deviation record of the lender, and generating the overdue behavior dynamic analysis result also includes:

[0032] S301: Collect credit score changes, new loan application times, and current overdue frequency data through the credit dynamic deviation records of the lender, screen abnormal credit behaviors of the lender under economic pressure, and obtain an overview of the credit activities of the lender under economic pressure;

[0033] S302: Based on the credit activity overview of the borrower under the economic pressure, analyze the borrower's consumption pattern, including abnormal expenditure increase or credit card usage frequency increase, and multiple change points, identify and mark the behavior trend leading to overdue payment through the change points, and obtain key behavior change records;

[0034] S303: Using the key behavior change records, assess the borrower's likelihood of default under various economic pressures, distinguish the relationship between behavior preferences and default risks, and obtain dynamic analysis results of default behavior.

[0035] Preferably, the step of using the overdue behavior dynamic analysis results to construct a prediction model, predict the lender's overdue probability, and obtain the lender's overdue risk analysis results also includes:

[0036] S401: Generate a comprehensive data set of overdue behavior and credit activity by collating the dynamic analysis results of the overdue behavior, including credit activity deviations and overdue behavior trends under economic pressure;

[0037] S402: Based on the comprehensive dataset of overdue behavior and credit activities, the behavior data of the lender is analyzed in combination with the cumulative prospect theory analysis method to identify the behavior pattern associated with the overdue probability and construct an overdue probability prediction model;

[0038] S403: Utilizing the overdue probability prediction model, comprehensively referring to the credit activity deviation under economic pressure, calculating the overdue probability of the lender, and obtaining the overdue risk analysis result of the lender.

[0039] Preferably, the cumulative prospect theory analysis method is based on the improved formula III:

[0040]

[0041] Calculate the lender's behavior data to obtain the lender's overdue probability value, and build an overdue probability prediction model based on it;

[0042] Where V′(B) represents the improved cumulative prospect value, i represents the index of the behavior, gains refers to the set of all behaviors that lead to gains, losses refers to the set of all behaviors that lead to losses, π′(p i ,θ) represents the improved weighting function, p i is the probability of behavior i, θ reflects the impact of market volatility or the dynamics of the credit market on probability perception, v′(x i , α, β, λ) represents the improved value function, x i represents the change in credit score or economic gain or loss caused by behavior i, p i represents the probability of behavior i occurring.

[0043] Preferably, the steps of classifying the lender risk levels according to the lender overdue risk analysis results, analyzing the key factors leading to the change in overdue probability, and generating risk classification and overdue factor identification results also include:

[0044] S501: classifying the risk levels of lenders by referring to the overdue probability value based on the overdue risk analysis results of the lenders, and obtaining a risk level list;

[0045] S502: Based on the risk level list, analyze the common characteristics and behaviors of lenders at each risk level, identify multiple key factors that affect the change in overdue probability, including income level, employment stability and debt ratio, and generate a list of key overdue factors;

[0046] S503: Using the key overdue factor list, perform iterative analysis on the lender groups of each risk level to reveal the correlation between each type of factors and obtain risk classification and overdue factor identification results.

[0047] Preferably, the step of analyzing the correlation of overdue factors in the risk classification and overdue factor identification results, identifying the interaction between each factor and its impact on overdue risk, and obtaining the risk factor correlation analysis results also includes:

[0048] S601: sorting the data of overdue factors in the risk classification and overdue factor identification results, collecting and classifying multiple key data on overdue probability, economic factors and personal behavior characteristics, and generating an overdue factor data set;

[0049] S602: Match and compare multiple data of the overdue factor data set to reveal the interaction between each factor of income level, employment stability and debt ratio, and generate an overdue factor correlation analysis table;

[0050] S603: Based on the overdue factor correlation analysis table, a correlation score is performed to quantify the impact intensity and action mode between the evaluation factors to obtain the risk factor correlation analysis results.

[0051] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0052] By using the Gaussian mixture model for dynamic clustering and combining it with the cumulative prospect theory to optimize the risk prediction model, the credit scoring model is made highly adaptable and dynamic, so that it can timely reflect changes in the market and personal financial status, ensuring that the borrower's credit rating accurately reflects its current credit status in real time. At the same time, by deeply analyzing the behavior of borrowers in different financial scenarios, the sensitivity and accuracy of overdue risk prediction are improved. For example, under economic pressure, the sudden change in the repayment behavior of borrowers can be identified and analyzed in a timely manner, allowing financial institutions to take corresponding measures before the borrower begins to default.

[0053] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 The present invention proposes a schematic diagram of the overall process of a method for intelligent identification and analysis of financial overdue risks;

[0055] Figure 2 The present invention proposes a specific flow chart of S1 of a method for intelligent identification and analysis of financial overdue risks;

[0056] Figure 3 The present invention proposes a specific flow chart of S2 of a method for intelligent identification and analysis of financial overdue risks;

[0057] Figure 4 The present invention proposes a specific flow chart of S3 of a method for intelligent identification and analysis of financial overdue risks;

[0058] Figure 5 The present invention proposes a specific flow chart of S4 of a method for intelligent identification and analysis of financial overdue risks;

[0059] Figure 6 The present invention proposes a specific flow chart of S5 of a method for intelligent identification and analysis of financial overdue risks;

[0060] Figure 7 The present invention proposes a specific flow chart of S6 of a method for intelligent identification and analysis of financial overdue risks. DETAILED DESCRIPTION

[0061] The present application provides a method for intelligently identifying and analyzing financial overdue risks.

[0062] Application Overview

[0063] There are problems in the existing technology. Although the existing technology has made significant progress in dealing with financial risk management, when the financial situation or market conditions of the borrower change, the traditional static credit scoring model is difficult to adjust the credit rating of the borrower in real time, which easily leads to the credit score being out of sync with the borrower's current real credit situation, affecting the risk management and decision-making efficiency of financial institutions. In addition, when evaluating the behavioral responses of borrowers in different financial scenarios, traditional models are difficult to accurately capture the nonlinear characteristics and loss aversion behaviors in the decision-making process. For example, when a borrower who repays the loan on time, after the economic pressure accumulates to a certain extent, he may suddenly start to delay or stop repayment. At the same time, the borrower may excessively reduce consumption and investment when facing potential financial difficulties, which affects the application effect and accuracy of the risk prediction model in actual financial decision-making.

[0064] In response to the above technical problems, the overall idea of ​​the technical solution provided by this application is as follows:

[0065] like Figure 1 As shown, the present application provides a method for intelligent identification and analysis of financial overdue risks, wherein the method includes:

[0066] S1: By clustering the lender credit data, identify lender groups with multiple credit ratings, determine the number of clusters using evaluation indicators, and generate a lender credit rating model;

[0067] S2: Based on the lender credit grading model, track the lender's credit activity data, capture changes in credit activities and record key events, and generate a record of the lender's credit dynamic deviation;

[0068] S3: Through the credit dynamic deviation records of lenders, analyze the behavior preference data of lenders under economic pressure, as well as the behavior trends that cause overdue payments, and generate overdue behavior dynamic analysis results;

[0069] S4: Using the results of dynamic analysis of overdue behavior, a prediction model is constructed to predict the overdue probability of the lender and obtain the overdue risk analysis results of the lender;

[0070] S5: Based on the overdue risk analysis results of the lenders, the lenders’ risk levels are divided, and the key factors leading to changes in overdue probability are analyzed to generate risk classification and overdue factor identification results;

[0071] S6: Analyze the correlation between overdue factors in the risk classification and overdue factor identification results, identify the interaction between each factor and its impact on overdue risk, and obtain the risk factor correlation analysis results.

[0072] The lender credit grading model includes credit score ranges, overdue record classification and financial health index. The lender credit dynamic deviation records include credit score decline events, number of new loan applications and overdue frequency. The overdue behavior dynamic analysis results include changes in consumption patterns under economic pressure, abnormal expenditure increases and credit card usage frequency. The lender overdue risk analysis results include the overdue probability value of each lender, the predicted time period for overdue occurrence and the overdue risk level classification. The risk classification and overdue factor identification results include lender groups graded by overdue probability, overdue economic factors and personal behavioral characteristics. The risk factor correlation analysis results include the correlation strength between financial status and personal behavioral characteristics, the interaction pattern between multiple economic factors and the analysis of complex factors affecting overdue risk.

[0073] like Figure 2 As shown, the steps of performing cluster analysis on the lender credit data, identifying lender groups with multiple credit ratings, determining the number of clusters using evaluation indicators, and generating a lender credit rating model also include:

[0074] S101: The specific process of clustering the lender's credit data, monitoring repayment behavior and account information through Gaussian mixture model and expectation maximization algorithm, identifying credit behavior patterns, determining the distribution quantity through Bayesian information criterion, and generating behavior pattern classification results is as follows;

[0075] The S101 sub-step estimates the model parameters based on the lender credit data through the Gaussian mixture model using the expectation-maximization algorithm, selects the Bayesian information criterion to compare the complexity of different models, identifies the optimal number of distributions, and uses the GaussianMixture function from the sklearn.mixture library by monitoring repayment behavior and account information. It specifies n_components as the number of clusters obtained by dynamic evaluation, sets the covariance_type parameter to full to allow each component to have its own general covariance matrix, sets max_iter to 200 to ensure algorithm convergence, initializes n_init to 10 to perform 10 independent runs with initialization, selects the best output, and generates the behavior pattern classification results.

[0076] Gaussian mixture model and expectation maximization algorithm are based on the improved formula I:

[0077]

[0078] Segment and identify financial overdue risks by referring to lender characteristics and credit rating information;

[0079] Among them, Q'(θ|θ (t) ) is given the current parameter estimate θ (t)After that, the improved conditional expected log-likelihood function represents the goodness of fit of the model under the current parameters. ij Represents the weight coefficient of the i-th data point under the j-th Gaussian distribution, taking into account the uneven impact of different lender characteristics on credit behavior patterns. ij ) is the posterior probability that the i-th data point comes from the j-th Gaussian distribution, indicating the probability that the data point belongs to this distribution. α, β, δ are coefficients used to adjust the model complexity and the influence of data features, allowing the model to adapt to different characteristics of the data more flexibly. α is evaluated by different mixing coefficients π j The impact on model performance is determined by considering the covariance matrix ∑ j The complexity of the model determines the impact of δ by evaluating the data point x i and the mean μ of each Gaussian distribution j The influence of the distance between them on the classification results is determined. j is the mixing coefficient of the jth distribution, which represents the proportion of this Gaussian distribution in the entire distribution. j is the covariance matrix of the jth Gaussian distribution, describing the correlation between variables in the distribution. i is a data point representing the observed credit data of the lender. j is the mean of the jth Gaussian distribution, representing the location of the center point of the distribution. k is the number of distributions, indicating the total number of Gaussian distributions assumed in the model. n is the number of data points, indicating how many lenders’ credit data are analyzed in total. i , C) is a data point x i The additional term L(x) and its credit rating C introduce additional information into the model and improve the accuracy of prediction. λ is the additional term coefficient, which adjusts the intensity of the impact of credit rating information on the final model. i , C) Contribution to model accuracy is determined.

[0080] The Bayesian Information Criterion follows Formula II:

[0081] BIC=-2ln(L)+kln(n)

[0082] Determine the optimal number of distributions and generate behavioral pattern classification results.

[0083] Among them, L is the maximum value of the likelihood function of the model, that is, the likelihood function value when the model parameters are optimized under given data. It reflects the degree of fit of the model to the observed data. The larger the likelihood function, the higher the probability that the data will appear under the current model and parameters. k is the number of parameters in the model. Including all parameters that need to be estimated. For example, in the Gaussian mixture model, the mean, covariance matrix, and mixing coefficient of each Gaussian distribution need to be estimated and all counted into k. n is the number of data points, that is, the sample size. It represents the total number of all lender records. ln represents the natural logarithm function. BIC penalizes the likelihood function value of the model. The penalty term depends on the number of parameters in the model and the number of data points. The purpose is to avoid overfitting. The more parameters in the model or the larger the amount of data, the larger the penalty term of BIC.

[0084] The execution process is as follows:

[0085] Collect the credit data of lenders, including repayment behavior, account information, etc. Pre-process the data, such as cleaning and standardization, to ensure the quality of input data.

[0086] Choose an initial number of distributions k to start with. For example, you can try different values ​​of k from small to large. For each Gaussian distribution parameter (mean μ j , covariance matrix∑ j ) and the mixing coefficient π j Make a preliminary estimate.

[0087] The E-step (expectation step) calculates the probability (posterior probability) γ(z) of each data point belonging to each Gaussian distribution by using the current model parameters. ij )0.

[0088] The M-step (maximization step) updates the model parameters, including the mean μ of each Gaussian distribution j , covariance matrix∑ j and the mixing coefficient π j , maximize the expected log-likelihood function calculated by the E-step.

[0089] Add the parameters in formula I to E-step and M-step, including the weight coefficient w ij , adjustment coefficients α, β, δ, and the additional term L(x i , C) and its coefficient λ.

[0090] For each value of k, the BIC value is calculated according to Formula II using the improved model and parameters, where L is the maximum value of the likelihood function given the data and model parameters, k is the number of parameters in the model, and n is the number of data points.

[0091] For different k values, the corresponding BIC values ​​are compared. The k with the smallest BIC value is selected as the optimal number of distributions, because a smaller BIC value indicates that the model has a better fit to the data while considering the complexity of the model.

[0092] Using the optimal k value and the corresponding model parameters, the data points are assigned to the most likely Gaussian distribution to complete the classification of the lender's credit behavior pattern.

[0093] S102: Based on the behavior pattern classification results, the repayment frequency and loan amount usage are analyzed, multiple credit rating features including payment behavior, account age and credit utilization are identified, and the specific process of generating credit rating difference analysis results is as follows;

[0094] In sub-step S102, based on the behavioral pattern classification results, logistic regression analysis is performed using the Logit function of the statsmodels library to extract features such as repayment frequency and loan amount usage as independent variables. Dummy variables are set up to process categorical data, such as payment behavior, account age, and credit utilization. C (payment behavior) + C (account age) + credit utilization is set as the model formula, the fit method is called for model training, and the predict method is used to predict credit rating differences based on the trained model. The output threshold is set to 0.5 to distinguish between high and low credit ratings, and the credit rating difference analysis results are generated.

[0095] S103: Using the credit rating difference analysis results, adjusting the classification standards, optimizing the credit rating group division, and generating the lender credit rating model. The specific process is:

[0096] In sub-step S103, based on the results of the credit rating difference analysis, the K-means clustering algorithm is adopted, the KMeans function of the scikit-learn library is used, n_clusters is specified as the number of groups determined by the credit rating difference analysis, the init parameter selects k-means++ to optimize the selection of the initial cluster center, max_iter is set to 300 to ensure the convergence of the algorithm, n_init is 10, 10 independent runs are performed to select the best result, the fit_predict method is called to cluster the credit data, and the classification criteria are dynamically adjusted by calculating the average distance from the data points in each group to its cluster center to generate a lender credit rating model.

[0097] like Figure 3 As shown, based on the lender credit grading model, the steps of tracking the lender credit activity data, capturing credit activity changes and recording key events, and generating the lender credit dynamic deviation record also include:

[0098] S201: Using the lender credit grading model, monitor the daily credit account activities of the lender, including current loan applications and repayment behaviors, and generate daily credit activity records in the following specific process;

[0099] Based on the lender credit rating model, sub-step S201 uses SQL query statements to retrieve the lender's daily credit account activities from the database, including SELECT statements for the loan application form and repayment record table, WHERE clause sets the date range to the current day, AND condition filters records with active account status, JOIN operation connects the user basic information table to obtain complete lender identity information, ORDER BY clause sorts the results in chronological order, uses Python's pandas library to convert the query results into a data frame, annotates each record with a timestamp and activity type, including the current loan application status and repayment behavior, and generates daily credit activity records.

[0100] S202: Screening abnormal activities from daily credit activity records, including credit score decline or new overdue records, and marking them as key credit change events. The specific process of generating key credit event records is as follows;

[0101] Sub-step S202 uses Python scripts to implement anomaly detection based on daily credit activity records. It defines threshold variables to set the sensitivity of credit score decline and identification rules for overdue records, and uses if-else logic to determine whether each record meets the conditions for abnormal activity. For cases where the credit score declines or new overdue records are added, the append method is called to add the record to a new list, assign a unique identifier to each abnormal activity, use the datetime library to mark the time of occurrence, and generate key credit event records.

[0102] S203: Analyze key credit event records, sort out the changing trends and key deviation points of credit activities, evaluate the health of credit activities, and generate the specific process of the lender's credit dynamic deviation record;

[0103] Based on the key credit event records, sub-step S203 uses data visualization tools such as Matplotlib and Seaborn library to draw a credit activity change trend chart. The chart type is set to a time series line chart. The X-axis represents time, and the Y-axis represents key credit activity indicators, including credit scores and the number of overdue records. The plot function is used to draw the change trend of each activity, and the scatter function is used to highlight the key deviation points. The describe method is called to perform statistical analysis on the key events, including the mean, median, and standard deviation. By comparing the statistical data and chart analysis of key indicators, the change trends and key deviation points of credit activities are sorted out to generate the lender's credit dynamic deviation record.

[0104] like Figure 4 As shown, the steps of analyzing the behavior preference data of the lender under economic pressure and the behavior trend causing overdue payment through the credit dynamic deviation record of the lender, and generating the overdue behavior dynamic analysis result also include:

[0105] S301: Through the credit dynamic deviation records of the lenders, the credit score changes, the number of new loan applications, and the current overdue frequency data are collected to screen the abnormal credit behaviors of the lenders under economic pressure, and the specific process of obtaining an overview of the credit activities of the lenders under economic pressure is as follows;

[0106] Sub-step S301 performs data collection based on the dynamic deviation records of the lender's credit, including the extraction of credit score changes, the number of new loan applications, and the current overdue frequency data. It uses data mining technology to extract the required fields from the database through SQL queries, and uses Python's Pandas library for data cleaning, including missing value processing and data type conversion. It screens out records of significant declines in credit scores, sharp increases in loan applications, and increases in overdue frequencies as abnormal credit behaviors, and screens out data points that fall within the abnormal range by setting thresholds. It further analyzes the changing trends of credit scores and overdue frequencies by drawing time series graphs, and obtains an overview of the lender's credit activities under economic pressure.

[0107] S302: Based on the credit activity overview of the borrower under economic pressure, analyze the borrower's consumption pattern, including abnormal increase in expenditure or increase in credit card usage frequency, and identify and mark the behavior trend that leads to overdue payment through the change time points. The specific process of obtaining the key behavior change record is as follows;

[0108] Sub-step S302 performs an analysis of the borrower's consumption patterns based on an overview of the borrower's credit activities under economic pressure. It adopts a time series analysis method and uses the ARIMA model in Python's statsmodels library to model the consumption data. The parameters are set to automatically identify the optimal difference order d, autoregressive term p, and moving average term q. The ACF and PACF graphs are used to assist in determining the parameter range, identify the time points when abnormal expenditures increase or the frequency of credit card use increases, and compare the model prediction results with the actual data. The behavioral trends that lead to overdue payments are identified through difference analysis, and a record of key behavioral changes is obtained.

[0109] S303: Using key behavior change records to assess the borrower's likelihood of default under various economic pressures, distinguishing the relationship between behavioral preferences and default risks, and obtaining the dynamic analysis results of default behavior are as follows;

[0110] The S303 sub-step uses key behavioral change records to perform overdue possibility assessment, adopts machine learning classification algorithms, implements the random forest algorithm through Python's scikit-learn library, based on the borrower's consumption pattern and credit behavior as feature input, and overdue or not as the target variable, sets n_estimators to 100 to build a forest of 100 trees, and sets max_depth to None to allow the tree depth to grow until all leaves are pure or contain leaves with less than min_samples_split samples. The train_test_split function is used to divide the data into training and test sets, with a ratio of 70% training and 30% testing. The fit method is used to train the model on the training data, and the predict method is used to predict the overdue situation of the test set. The matching degree between the evaluation results and the actual overdue records is further analyzed, and feature importance analysis is performed to distinguish the relationship between behavioral preferences and overdue risks, and the dynamic analysis results of overdue behavior are obtained.

[0111] like Figure 5 As shown, the steps of using the overdue behavior dynamic analysis results to construct a prediction model, predict the lender's overdue probability, and obtain the lender's overdue risk analysis results also include:

[0112] S401: The specific process of generating a comprehensive data set of overdue behavior and credit activity by collating the results of the dynamic analysis of overdue behavior, including the credit activity deviation and overdue behavior trend under economic pressure, is as follows;

[0113] Based on the results of the dynamic analysis of overdue behavior, sub-step S401 uses Python language combined with the Pandas library to organize the data, aggregates the credit activity deviation and overdue behavior trend under economic pressure through the DataFrame object, applies the groupby method to group each lender by month, uses the agg function to calculate the average credit score change and overdue event count of each group, and merges the economic indicator data through the merge method to generate a comprehensive data set of overdue behavior and credit activity.

[0114] S402: Based on the comprehensive data set of overdue behavior and credit activities, the specific process of analyzing the behavior data of lenders in combination with the cumulative prospect theory analysis method, identifying the behavior patterns associated with the overdue probability, and constructing the overdue probability prediction model is as follows;

[0115] Sub-step S402 is based on the comprehensive dataset of overdue behavior and credit activities, and uses the cumulative prospect theory analysis method to analyze. The logistic regression model is constructed using the Python language combined with the Scikit-learn library. The model is initialized through the LogisticRegression class, and the solver parameters are set to liblinear to adapt to the scale and characteristics of the dataset. The fit method is used to input behavioral data and overdue labels for model training, identify the association between overdue probability and behavioral patterns, and construct an overdue probability prediction model.

[0116] The cumulative prospect theory analysis method is based on the improved formula III:

[0117]

[0118] Calculate the lender's behavior data to obtain the lender's overdue probability value, and build an overdue probability prediction model based on it;

[0119] Where V′(B) represents the improved cumulative outlook value, reflecting the expected value of overdue risk of the lender under economic pressure. i represents the index of the behavior, which is used to traverse all possible behaviors or events. Gains refers to the set of all behaviors that lead to gains. Losses refers to the set of all behaviors that lead to losses. π′(p i ,θ) represents the improved weighting function, p i is the probability of occurrence of behavior i, θ is a newly introduced parameter that reflects the impact of market volatility or the dynamics of the credit market on probability perception, and is determined by the market volatility index or the historical data of the credit market overdue rate. i , α, β, λ) represents the improved value function, where x i is the change in credit score or economic gain or loss caused by behavior i, α and β are parameters for adjusting gain and loss sensitivity, respectively determined by analyzing the sensitivity of lenders’ credit behavior under different economic conditions and the intensity of lenders’ response to losses, and λ is a dynamic adjustment parameter of the loss aversion coefficient, which is determined dynamically by evaluating lenders’ loss aversion behavior and combining market trend data. i Represents the credit score change or economic gain or loss caused by behavior i. i represents the probability of behavior i occurring.

[0120] The execution process is as follows:

[0121] Collect data on lenders’ credit activity, including credit activity deviations and delinquency behavior trends under economic stress, as well as market volatility or credit market dynamics.

[0122] The parameter θ in the weighting function is determined using the market volatility index or historical data on overdue rates to reflect the impact of market dynamics on probability perception.

[0123] The parameters α, β and λ in the value function are dynamically adjusted according to the lender's historical behavior data and market conditions. The parameters adjust the return sensitivity, loss sensitivity and loss aversion respectively, accurately reflecting the behavior patterns of lenders under different economic conditions.

[0124] For each behavior or event i, calculate the improved weighting function π′(p i ,θ), the original occurrence probability p i By adjusting the parameter θ, the impact of market fluctuations or other external factors can be considered.

[0125] For each action or event i, calculate the improved value function v′(x i , α, β, λ). Changes in credit scores or economic gains and losses x caused by the behavior i . And the parameters α, β, λ calculate the value, reflecting the lender's sensitivity to gains and losses and loss aversion.

[0126] Through formula III, the products of the improved value function and the weighted function in all gain cases are accumulated, and then the corresponding products in all loss cases are accumulated, and finally the two results are added together to obtain the improved cumulative prospect value V′(B).

[0127] The improved cumulative prospect value V′(B) is used to predict the overdue probability of the lender. In the model, V′(B) can be used to evaluate the overdue risk under different behavior patterns, thus providing support for risk management and decision-making.

[0128] S403: Using the overdue probability prediction model, comprehensively referring to the credit activity deviation under economic pressure, calculating the overdue probability of the lender, and obtaining the overdue risk analysis result of the lender is as follows;

[0129] Sub-step S403 uses the overdue probability prediction model, and uses Python language and Numpy library to digitize the credit activity deviation under economic pressure. The overdue probability of the lender is calculated based on the trained logistic regression model through the predict_proba method. For each lender, the argmax function is used to select the highest overdue risk level from the probability results to generate the lender's overdue risk analysis results.

[0130] like Figure 6 As shown, according to the overdue risk analysis results of the lenders, the risk levels of the lenders are divided, and the key factors leading to the change of overdue probability are analyzed to generate the risk classification and overdue factor identification results, which also includes:

[0131] S501: Based on the overdue risk analysis results of the lenders, the lenders' risk levels are classified with reference to the overdue probability values, and the specific process of obtaining the risk level list is as follows;

[0132] Sub-step S501 performs risk level classification based on the results of the lender overdue risk analysis. It adopts the threshold segmentation method, uses Python scripts to process data through the Pandas library, sets the numerical range of overdue probability as the threshold, and classifies lenders with an overdue probability below 5% as low risk, lenders with an overdue probability between 5% and 20% as medium risk, and lenders with an overdue probability above 20% as high risk. The cut function is used to map the overdue probability value to the corresponding risk level, generate a DataFrame containing the lender ID and the corresponding risk level, and generate a risk level list.

[0133] S502: Based on the risk level list, analyze the common characteristics and behaviors of lenders at each risk level, identify multiple key factors that affect the change in the probability of overdue payment, including income level, employment stability and debt ratio, and generate a list of key overdue factors in the following specific process;

[0134] Sub-step S502 performs feature and behavior analysis based on the risk level list, adopts multivariate analysis methods, and performs generalized linear model analysis through the glm function of R language. It takes income level, employment stability and debt ratio as independent variables, and risk level as dependent variable, sets the family parameter to binomial for logistic regression, and identifies the common features and behaviors of borrowers with different risk levels. The anova function performs variance analysis on the model to test the influence of each variable, identifies factors that significantly affect the change in the probability of overdue payment, and generates a list of key overdue factors.

[0135] S503: Using the key overdue factor list, iteratively analyzing the lender groups of each risk level, revealing the correlation between each type of factors, and obtaining the risk classification and overdue factor identification results. The specific process is as follows;

[0136] Sub-step S503 uses the list of key overdue factors to perform iterative analysis, adopts cluster analysis, and uses the KMeans algorithm in Python's scikit-learn library to set the n_clusters parameter to the number of different risk levels in the lender group to perform internal segmentation of the lender group. The fit_predict method performs cluster analysis on the lender group of each risk level based on income level, employment stability, and debt ratio, revealing the relationship between different key overdue factors and the difference in their impact on overdue risk within the same risk level, and obtaining risk classification and overdue factor identification results.

[0137] like Figure 7As shown, the steps of analyzing the correlation of overdue factors in the risk classification and overdue factor identification results, identifying the interaction between each factor and its impact on overdue risk, and obtaining the risk factor correlation analysis results also include:

[0138] S601: sorting the data of overdue factors in the risk classification and overdue factor identification results, collecting and classifying multiple key data on overdue probability, economic factors and personal behavior characteristics, and generating the overdue factor data set in the following specific process;

[0139] Sub-step S601 performs data collation based on the risk classification and overdue factor identification results, adopts data aggregation technology, uses Python scripts to operate on the data through the Pandas library, uses the groupby function to group the data by risk level, and uses the agg function to calculate the average value of the overdue probability, economic factors such as the median income level, and personal behavioral characteristics such as the average debt ratio under each risk level, classifies and integrates the data, and merges the results into a DataFrame to provide a basis for subsequent analysis and generate a data set of overdue factors.

[0140] S602: Match and compare multiple data of the overdue factor data set to reveal the interaction between each factor of income level, employment stability and debt ratio, and generate the overdue factor correlation analysis table. The specific process is as follows;

[0141] Sub-step S602 performs correlation analysis between factors based on the overdue factor dataset, adopts the Pearson correlation coefficient calculation method, uses the corrcoef function through Python's NumPy library to compare factors such as income level, employment stability and debt ratio, and calculates the correlation coefficient. It draws a heat map through the matplotlib library to visualize the correlation between factors, revealing how the interaction between different economic factors and personal behavioral characteristics affects the overdue probability, and generates an overdue factor correlation analysis table.

[0142] S603: Based on the overdue factor correlation analysis table, the correlation score is performed to quantify the impact intensity and action mode between the evaluation factors, and the specific process of obtaining the risk factor correlation analysis result is as follows;

[0143] Sub-step S603 performs correlation scoring based on the overdue factor correlation analysis table. It adopts linear regression analysis and uses the lm function in R language to model the linear relationship between the overdue probability and various factors. It sets the overdue probability as the dependent variable, and the income level, employment stability and debt ratio as independent variables. It obtains the coefficient estimates and p-values ​​of the model through the summary function, quantitatively evaluates the impact intensity and action mode of each factor on the overdue probability, performs diagnostic checks on the model to ensure that there are no multicollinearity problems, obtains the correlation score of each factor on the overdue risk, and generates the risk factor correlation analysis results.

[0144] Although the present application has been described in conjunction with specific features and embodiments thereof, it is obvious that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and the drawings are merely exemplary illustrations of the present application as defined by the appended claims, and are deemed to have covered any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include these modifications and variations.

Claims

1. A method for intelligent identification and analysis of financial overdue risk, characterized in that: The method comprises: By clustering the lender credit data, we can identify lender groups with multiple credit ratings, determine the number of clusters using evaluation indicators, and generate a lender credit rating model; Based on the lender credit grading model, track the lender's credit activity data, capture changes in credit activities and record key events, and generate a dynamic deviation record of the lender's credit; By using the credit dynamic deviation records of the lender, the behavior preference data of the lender under economic pressure and the behavior trends that cause overdue payments are analyzed to generate overdue behavior dynamic analysis results; Using the dynamic analysis results of overdue behavior, a prediction model is constructed to predict the overdue probability of the lender and obtain the overdue risk analysis results of the lender; According to the overdue risk analysis results of the lenders, the lenders’ risk levels are divided, and the key factors leading to changes in overdue probability are analyzed to generate risk classification and overdue factor identification results; The correlation between the overdue factors in the risk classification and overdue factor identification results is analyzed to identify the interaction between each factor and its impact on the overdue risk, and obtain the risk factor correlation analysis results.

2. The method for intelligent identification and analysis of financial overdue risk according to claim 1, characterized in that: The lender credit grading model includes credit score ranges, overdue record classifications and financial health indexes; the lender credit dynamic deviation records include credit score decline events, the number of new loan applications and the frequency of overdue occurrences; the overdue behavior dynamic analysis results include changes in consumption patterns under economic pressure, increases in abnormal expenditures and the frequency of credit card use; the lender overdue risk analysis results include the overdue probability value of each lender, the predicted time period for overdue occurrence and the overdue risk level classification; the risk classification and overdue factor identification results include lender groups graded by overdue probability, overdue economic factors and personal behavioral characteristics; the risk factor correlation analysis results include the correlation strength between financial status and personal behavioral characteristics, the interaction pattern between multiple economic factors and the analysis of complex factors affecting overdue risks.

3. The method for intelligent identification and analysis of financial overdue risk according to claim 1, characterized in that: The steps of performing cluster analysis on the lender credit data, identifying lender groups with multiple credit ratings, determining the number of clusters using evaluation indicators, and generating a lender credit rating model also include: By clustering the lender's credit data, using the Gaussian mixture model, and using the expectation maximization algorithm to monitor repayment behavior and account information, identify credit behavior patterns, determine the number of distributions through the Bayesian information criterion, and generate behavior pattern classification results; Based on the behavioral pattern classification results, analyze the repayment frequency and loan amount usage, identify multiple credit rating characteristics including payment behavior, account age and credit utilization, and generate credit rating difference analysis results; The credit rating difference analysis results are used to adjust the classification standards, optimize the credit rating group division, and generate a lender credit grading model.

4. The method for intelligent identification and analysis of financial overdue risk according to claim 3 is characterized in that: The Gaussian mixture model and expectation maximization algorithm are based on the improved formula I: Segment and identify financial overdue risks by referring to lender characteristics and credit rating information; Among them, Q'(θ|θ (t) ) is given the current parameter estimate θ (t) After that, the improved conditional expected log-likelihood function, w ij represents the weight coefficient of the i-th data point under the j-th Gaussian distribution, γ(z ij ) is the posterior probability that the i-th data point comes from the j-th Gaussian distribution, α, β, δ are coefficients used to adjust the model complexity and the influence of data features, π j is the mixing coefficient of the jth distribution, ∑ j is the covariance matrix of the jth Gaussian distribution, x i is a data point, μ j is the mean of the jth Gaussian distribution, k is the number of distributions, n is the number of data points, L(x i , C) is a i and the additional term of its credit rating C, λ is the additional term coefficient; The Bayesian Information Criterion follows Formula II: BIC=-2ln(L)+kln(n) Determine the optimal distribution quantity and generate behavior pattern classification results; Among them, L is the maximum value of the likelihood function of the model, k is the number of parameters in the model, n is the number of data points, ln represents the natural logarithm function, and BIC penalizes the value of the likelihood function of the model.

5. The method for intelligent identification and analysis of financial overdue risk according to claim 1, characterized in that: Based on the lender credit grading model, the steps of tracking the lender credit activity data, capturing credit activity changes and recording key events, and generating a lender credit dynamic deviation record also include: Using the lender credit grading model, monitor the daily credit account activities of the lender, including current loan applications and repayment behaviors, and generate daily credit activity records; Screening abnormal activities from the daily credit activity records, including a decrease in credit score or a new overdue record, and marking them as key credit change events to generate key credit event records; Analyze the key credit event records, sort out the changing trends and key deviation points of credit activities, evaluate the health of credit activities, and generate dynamic deviation records of lenders' credit.

6. The method for intelligent identification and analysis of financial overdue risk according to claim 1, characterized in that: The step of analyzing the behavior preference data of the lender under economic pressure and the behavior trend causing overdue payment through the credit dynamic deviation record of the lender, and generating the overdue behavior dynamic analysis result also includes: By using the credit dynamic deviation records of the lenders, data on credit score changes, the number of new loan applications, and the current frequency of overdue payments are collected to screen for abnormal credit behaviors of lenders under economic pressure, and obtain an overview of the credit activities of lenders under economic pressure; Based on the credit activity overview of the borrower under the economic pressure, analyze the borrower's consumption pattern, including abnormal increase in expenditure or increase in credit card usage frequency, and identify and mark the behavioral trends that lead to overdue payments through the change points, and obtain key behavioral change records; The key behavior change records are used to evaluate the borrower's possibility of default under various economic pressures, distinguish the relationship between behavioral preferences and default risks, and obtain dynamic analysis results of default behavior.

7. The method for intelligent identification and analysis of financial overdue risk according to claim 1, characterized in that: The step of using the overdue behavior dynamic analysis results to construct a prediction model, predict the overdue probability of the lender, and obtain the overdue risk analysis results of the lender also includes: Generate a comprehensive dataset of overdue behavior and credit activity by collating the above-mentioned overdue behavior dynamic analysis results, including credit activity deviations and overdue behavior trends under economic pressure; Based on the above-mentioned comprehensive dataset of overdue behavior and credit activities, the behavior data of lenders is analyzed in combination with the cumulative prospect theory analysis method to identify the behavior patterns associated with overdue probability and build an overdue probability prediction model; The overdue probability prediction model is used to comprehensively refer to the credit activity deviation under economic pressure to calculate the overdue probability of the lender and obtain the overdue risk analysis result of the lender.

8. The method for intelligent identification and analysis of financial overdue risk according to claim 7, characterized in that: The cumulative prospect theory analysis method is based on the improved formula III: Calculate the lender's behavior data to obtain the lender's overdue probability value, and build an overdue probability prediction model based on it; Where V′(B) represents the improved cumulative prospect value, i represents the index of the behavior, gains refers to the set of all behaviors that lead to gains, losses refers to the set of all behaviors that lead to losses, π′(p i ,θ) represents the improved weighting function, p i is the probability of behavior i, θ reflects the impact of market volatility or the dynamics of the credit market on probability perception, v′(x i , α, β, λ) represents the improved value function, x i represents the change in credit score or economic gain or loss caused by behavior i, p i represents the probability of behavior i occurring.

Citation Information

Cited By

  • Loan consumption fund use monitoring method and device and electronic equipment

    CN120047237A