CN118941379A - A method for intelligent identification and analysis of financial overdue risks - Google Patents

Abstract

The invention relates to the technical field of financial risk management, in particular to an intelligent identification and analysis method for financial overdue risk, which comprises the following steps: by performing cluster analysis on the credit data of the borrowers, the credit groups of multiple types of credit grades are identified, the number of clusters is determined by using the evaluation index, and a credit grading model of the borrowers is generated. According to the invention, dynamic clustering is carried out through the Gaussian mixture model, and the risk prediction model is optimized by combining the accumulated prospect theory, so that the credit scoring model has high adaptability and dynamic property, thereby timely reflecting the changes of market and personal financial conditions and ensuring that the credit level of a borrower accurately reflects the current credit state of the borrower in real time. Meanwhile, by deeply analyzing the behaviors of the borrower under different financial situations, the sensitivity and accuracy of overdue risk prediction are improved, and the financial institution is allowed to take corresponding measures before the borrower starts overdue.

Description

Intelligent identification and analysis method for financial overdue risk

Technical Field

The invention relates to the technical field of financial risk management, in particular to an intelligent identification and analysis method for financial overdue risks.

Background

The financial risk management technology field includes methods and tools for identifying, evaluating, monitoring and mitigating various risks in the financial market. The field adopts data analysis technology, artificial intelligence, machine learning and big data technology, and improves the accuracy and efficiency of risk management. Enabling the financial institution to better predict and guard against potential risk events.

The method for intelligently identifying and analyzing the overdue risk of the finance is a specific practice in the technical field of financial risk management, and focuses on identifying and analyzing the overdue risk of a finance product (such as loan and credit card) by using an intelligent technology. The objective is to help financial institutions take precautionary measures by early identification of potential default or overdue payment risk, thereby reducing losses and optimizing asset quality.

Although the prior art has made significant progress in dealing with financial risk management, when the financial situation or market condition of a borrower changes, it is difficult for a conventional static credit rating model to adjust the credit rating of the borrower in real time, which easily results in a credit rating that is not synchronized with the current real credit situation of the borrower, affecting the risk management and decision efficiency of the financial institution. In addition, when evaluating behavioral responses of a borrower in the face of different financial scenarios, it is difficult for a conventional model to accurately capture nonlinear characteristics and loss aversion behavior in the decision process, for example, when a borrower repayment on time, after economic pressure builds up to a certain extent, repayment may suddenly start to be delayed or stopped, and at the same time, the borrower may excessively reduce consumption and investment in the face of potential financial difficulties, affecting the application effect and accuracy of the risk prediction model in actual financial decisions.

Disclosure of Invention

The application provides an intelligent identification and analysis method for financial overdue risk, which solves the problems that although the prior art has significantly progressed in the aspect of processing financial risk management, when the financial situation or market condition of a borrower changes, the credit grade of the borrower is difficult to adjust in real time by a traditional static credit grading model, so that the credit grading is asynchronous with the current real credit situation of the borrower, and the risk management and decision-making efficiency of a financial institution are affected. In addition, when evaluating behavioral responses of a borrower in the face of different financial scenarios, it is difficult for a conventional model to accurately capture nonlinear characteristics and loss aversion behavior in the decision process, for example, when a borrower repayment on time, after economic pressure builds up to a certain extent, repayment may suddenly start to be delayed or stopped, and at the same time, the borrower may excessively reduce consumption and investment in the face of potential financial difficulties, which affects the application effect and accuracy of the risk prediction model in actual financial decisions.

In view of the above problems, the present application provides a financial overdue risk intelligent identification and analysis method.

The application provides an intelligent identification and analysis method for financial overdue risk, wherein the method comprises the following steps:

s1: identifying groups of borrowers with multiple credit classes by carrying out cluster analysis on the credit data of the borrowers, determining the cluster number by using the evaluation index, and generating a credit grading model of the borrowers;

S2: tracking the credit activity data of the borrower based on the credit grading model, capturing the credit activity change and recording key events to generate a credit dynamic deviation record of the borrower;

s3: analyzing behavior preference data of the borrower under economic pressure and behavior trend causing overdue through the credit dynamic deviation record of the borrower, and generating overdue behavior dynamic analysis results;

s4: constructing a prediction model by using the overdue behavior dynamic analysis result, predicting overdue probability of the borrower, and obtaining overdue risk analysis result of the borrower;

s5: dividing the risk level of the borrower according to the overdue risk analysis result of the borrower, and analyzing key factors causing overdue probability change to generate risk classification and overdue factor identification results;

s6: and analyzing the association of the overdue factors in the risk classification and overdue factor identification results, and identifying the interaction among each factor and the influence of each factor on overdue risk to obtain a risk factor association degree analysis result.

Preferably, the borrower credit rating model comprises a credit rating interval, a overdue record classification and a financial health index, the borrower credit dynamic deviation record comprises a credit score decreasing event, a new loan application number and a overdue occurrence frequency, the overdue behavior dynamic analysis result comprises a consumption mode change under economic pressure, an abnormal expenditure increase and a credit card use frequency, the borrower overdue risk analysis result comprises a overdue probability value, a time period for predicting overdue occurrence and overdue risk level division of each borrower, the risk classification and overdue factor identification result comprises a borrower group classified according to the overdue probability, overdue economic factors and personal behavior characteristics, and the risk factor association degree analysis result comprises an association strength of financial conditions and personal behavior characteristics, an interaction mode among multiple economic factors and a composite factor analysis for influencing overdue risks.

Preferably, the step of generating the credit rating model further comprises the steps of identifying a plurality of types of credit level groups of the borrower by performing cluster analysis on the credit data, determining the number of clusters by using the evaluation index, and:

S101: the credit data of the borrower is subjected to clustering analysis, the repayment behaviors and account information are monitored through a Gaussian mixture model by adopting an expectation maximization algorithm, a credit behavior mode is identified, the distribution quantity is determined through a Bayesian information quantity criterion, and a behavior mode classification result is generated;

s102: based on the behavior pattern classification result, analyzing repayment frequency and loan amount usage, identifying a plurality of credit level characteristics including payment behavior, account age and credit utilization rate, and generating a credit level difference analysis result;

S103: and adjusting the classification standard by utilizing the credit level difference analysis result, optimizing credit level group division, and generating a borrower credit grading model.

Preferably, the gaussian mixture model and the expectation maximization algorithm follow the improved formula I:

subdividing and identifying financial overdue risks by referring to the borrower characteristics and the credit rating information;

Where Q' (θ|θ ^(t)) is the log likelihood function expected for the improved condition given the current parameter estimate θ ^(t), w _ij represents the weight coefficient of the ith data point at the jth Gaussian distribution, γ (z _ij) is the posterior probability of the ith data point from the jth Gaussian distribution, α, β, δ is the coefficient for adjusting the model complexity and the data characteristic impact, pi _j is the mixing coefficient of the jth distribution, Σ _j is the covariance matrix of the jth Gaussian distribution, x _i is the 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 an additional term for data point x _i and its confidence level C, λ is the additional term coefficient;

The bayesian information criterion is according to formula II:

BIC＝-2ln(L)+kln(n)

determining the optimal distribution quantity and generating a behavior pattern classification result;

Where L is the maximum of the likelihood function of the model, k is the number of parameters in the model, n is the number of data points, ln is the natural logarithmic function, and BIC penalizes the likelihood function value of the model.

Preferably, the step of tracking the borrower credit activity data, capturing the credit activity change and recording the key event based on the borrower credit rating model, and generating a borrower credit dynamic deviation record further comprises:

S201: monitoring activities of a daily credit account of a borrower by adopting the borrower credit rating model, wherein the activities comprise current loan application and repayment behaviors, and generating a daily credit activity record;

S202: screening abnormal activities from the daily credit activity records, wherein the abnormal activities comprise credit score decrease or newly added overdue records, marking the abnormal activities as key credit change events, and generating key credit event records;

S203: and analyzing the key credit event records, sorting the variation trend and key deviation points of credit activities, evaluating the health condition of the credit activities, and generating a dynamic credit deviation record of a borrower.

Preferably, the step of analyzing the behavior preference data of the borrower under the economic pressure and the behavior trend causing overdue through the credit dynamic deviation record of the borrower, and generating the overdue behavior dynamic analysis result further comprises:

s301: collecting credit score variation, newly-added loan application times and current overdue frequency data through the dynamic deviation record of the borrower credit, screening abnormal credit behaviors of the borrower under economic pressure, and obtaining a credit activity overview of the borrower under economic pressure;

S302: based on the credit activity overview of the borrower under the economic pressure, analyzing a plurality of change time points of abnormal expenditure increase or credit card use frequency increase, and identifying and marking behavior trends causing overdue through the change time points to obtain key behavior change records;

S303: and evaluating overdue possibility of the borrower under various economic pressures by adopting the key behavior change record, distinguishing the relation between behavior preference and overdue risk, and obtaining the overdue behavior dynamic analysis result.

Preferably, the step of constructing a prediction model by using the overdue behavior dynamic analysis result to predict the overdue probability of the borrower and obtain the overdue risk analysis result of the borrower further comprises:

S401: generating a comprehensive data set of overdue behaviors and credit activities by arranging the overdue behavior dynamic analysis results, including credit activity deviation and overdue behavior trend under economic pressure;

S402: analyzing the behavior data of the borrower based on the overdue behavior and credit activity comprehensive data set by combining an accumulated prospect theory analysis method, identifying a behavior mode associated with overdue probability, and constructing an overdue probability prediction model;

S403: and calculating the overdue probability of the borrower by comprehensively referring to the credit activity deviation under the economic pressure by using the overdue probability prediction model, and obtaining the overdue risk analysis result of the borrower.

Preferably, the cumulative prospect theory analysis method is according to improved formula III:

Calculating behavior data of a borrower to obtain overdue probability values of the borrower, and constructing a overdue probability prediction model according to the overdue probability values;

Where V ' (B) represents the modified cumulative prospect value, i represents the index of behavior, gains represents the set of behavior that all results in revenue, losses represents the set of behavior that all results in loss, pi ' (p _i, θ) represents the modified weighting function, p _i is the probability of occurrence of behavior i, θ reflects the impact of market volatility or credit market dynamics on probability perception, V ' (x _i, α, β, λ) represents the modified cost function, x _i represents the credit score change or economic loss caused by behavior i, and p _i represents the probability of occurrence of behavior i.

Preferably, the steps of classifying the loan risk according to the result of the analysis of the overdue risk of the borrower, analyzing key factors causing the change of overdue probability, and generating risk classification and overdue factor identification result further comprise:

S501: dividing the risk level of the borrower according to the overdue risk analysis result of the borrower and referring to the overdue probability value to obtain a risk level list;

S502: analyzing common characteristics and behaviors of each risk level borrower based on the risk level list, identifying a plurality of key factors affecting overdue probability changes including income level, employment stability and liability ratio, and generating a key overdue factor list;

S503: and carrying out iterative analysis on the borrower groups of each risk level by utilizing the key overdue factor list, revealing the relevance among each type of factors, and obtaining risk classification and overdue factor identification results.

Preferably, the step of analyzing the risk classification and the correlation of the overdue factors in the overdue factor recognition result, and recognizing the interaction between each factor and the influence of each factor on the overdue risk to obtain a risk factor correlation analysis result further includes:

S601: data arrangement is carried out on overdue factors in the risk classification and overdue factor identification results, multiple key data of overdue probability, economic factors and personal behavior characteristics are collected and classified and integrated, and an overdue factor data set is generated;

S602: matching and comparing multiple data of the overdue factor data set to reveal the interaction among each factor of income level, employment stability and liability ratio, and generating an overdue factor correlation analysis table;

S603: and carrying out relevance scoring based on the overdue factor relevance analysis table, and quantifying influence intensity and action mode among evaluation factors to obtain a risk factor relevance analysis result.

One or more technical schemes provided by the application have at least the following technical effects or advantages:

Dynamic clustering is carried out through a Gaussian mixture model, and a risk prediction model is optimized by combining an accumulated prospect theory, so that a credit scoring model has high adaptability and dynamic property, changes of market and personal financial conditions can be timely reflected, and the credit level of a borrower can be ensured to accurately reflect the current credit state of the borrower in real time. Meanwhile, by deeply analyzing behaviors of the borrower under different financial situations, the sensitivity and accuracy of overdue risk prediction are improved. For example, under economic pressure, changes in the borrower's likely abrupt changes in repayment behavior can be timely identified and analyzed, allowing the financial institution to take corresponding action before the borrower has not yet begun overdue.

The foregoing description is only an overview of the present application, and is intended to be implemented in accordance with the teachings of the present application in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present application more readily apparent.

Drawings

FIG. 1 is a schematic diagram of an overall flow chart of a method for intelligently identifying and analyzing financial overdue risk according to the present invention;

FIG. 2 is a schematic diagram of an S1 embodiment of a method for intelligently identifying and analyzing financial overdue risk according to the present invention;

FIG. 3 is a schematic diagram of an S2 embodiment of a method for intelligently identifying and analyzing financial overdue risk according to the present invention;

FIG. 4 is a schematic diagram of a specific flow of S3 of a method for intelligently identifying and analyzing financial overdue risk according to the present invention;

FIG. 5 is a schematic diagram showing a specific flow of S4 of a method for intelligently identifying and analyzing financial overdue risk according to the present invention;

FIG. 6 is a schematic diagram showing an S5 embodiment of a method for intelligently identifying and analyzing financial overdue risk according to the present invention;

Fig. 7 is a schematic diagram of a specific flow of S6 of a method for intelligently identifying and analyzing financial overdue risk according to the present invention.

Detailed Description

The application provides an intelligent identification and analysis method for financial overdue risks.

Summary of the application

Although the prior art has made remarkable progress in dealing with financial risk management, when the financial situation or market condition of a borrower changes, the conventional static credit rating model is difficult to adjust the credit rating of the borrower in real time, which easily causes that the credit rating is not synchronous with the current real credit situation of the borrower, and affects the risk management and decision-making efficiency of a financial institution. In addition, when evaluating behavioral responses of a borrower in the face of different financial situations, it is difficult for the conventional model to accurately capture nonlinear characteristics and loss aversion behavior in the decision process, for example, when a borrower repayment on time, after economic pressure builds up to a certain extent, repayment may suddenly start to be delayed or stopped, and at the same time, the borrower may excessively reduce consumption and investment in the face of potential financial difficulties, which affects the technical problems of the application effect and accuracy of the risk prediction model in actual financial decisions.

Aiming at the technical problems, the technical scheme provided by the application has the following overall thought:

as shown in fig. 1, the application provides a financial overdue risk intelligent identification and analysis method, wherein the method comprises the following steps:

s1: identifying groups of borrowers with multiple credit classes by carrying out cluster analysis on the credit data of the borrowers, determining the cluster number by using the evaluation index, and generating a credit grading model of the borrowers;

s2: tracking the credit activity data of the borrower based on the credit grading model, capturing the credit activity change and recording key events to generate a credit dynamic deviation record of the borrower;

s3: analyzing behavior preference data of a borrower under economic pressure and behavior trend causing overdue through the credit dynamic deviation record of the borrower, and generating overdue behavior dynamic analysis results;

S4: constructing a prediction model by using the overdue behavior dynamic analysis result, and predicting overdue probability of the borrower to obtain overdue risk analysis result of the borrower;

S5: dividing the risk level of the borrower according to the overdue risk analysis result of the borrower, and analyzing key factors causing overdue probability change to generate risk classification and overdue factor identification results;

S6: and analyzing the association of overdue factors in the risk classification and overdue factor identification results, and identifying the interaction among each factor and the influence of each factor on overdue risk to obtain a risk factor association degree analysis result.

The borrower credit rating model comprises a credit rating interval, overdue record classification and a financial health index, the borrower credit dynamic deviation record comprises credit score decreasing events, new loan application times and overdue occurrence frequency, the overdue behavior dynamic analysis result comprises consumption mode change under economic pressure, abnormal expenditure increase and credit card use frequency, the borrower overdue risk analysis result comprises overdue probability values of each borrower, time period for predicting overdue occurrence and overdue risk level division, the risk classification and overdue factor identification result comprises a borrower group classified according to the overdue probability, overdue economic factors and personal behavior characteristics, and the risk factor association degree analysis result comprises association strength of financial conditions and personal behavior characteristics, interaction modes among multiple economic factors and compound factor analysis for influencing overdue risks.

As shown in fig. 2, the step of generating a credit rating model by performing cluster analysis on credit data, identifying a plurality of credit groups of credit classes, determining the number of clusters using an evaluation index, and further includes:

S101: the credit data of the borrower is subjected to cluster analysis, the repayment behavior and account information are monitored through a Gaussian mixture model by adopting an expectation maximization algorithm, a credit behavior mode is identified, the distribution quantity is determined through a Bayesian information quantity criterion, and the specific process of generating a behavior mode classification result is as follows;

The S101 substep estimates model parameters based on borrower credit data through a Gaussian mixture model, uses a expectation maximization algorithm to compare different model complexity, selects a Bayesian information criterion to identify optimal distribution quantity, monitors repayment behaviors and account information, utilizes GaussianMixture functions to designate n_components as clustering quantity obtained by dynamic evaluation from a sklearn. Mixture library, sets a covariance_type parameter to full to allow each component to have own general covariance matrix, sets max_iter to 200 to ensure algorithm convergence, initializes n_init to 10 to perform 10 independent operations of initialization, and selects optimal output to generate a behavior pattern classification result.

The gaussian mixture model and the expectation maximization algorithm follow the modified formula I:

subdividing and identifying financial overdue risks by referring to the borrower characteristics and the credit rating information;

Where Q' (θ|θ ^(t)) is a log-likelihood function expected from the improved condition given the current parameter estimate θ ^(t), representing the goodness of fit of the model under the current parameter. w _ij represents the weighting coefficient of the ith data point at the jth gaussian distribution, accounting for the imbalance of different borrower characteristics' effects on the credit behavior pattern. Gamma (z _ij) is the posterior probability that the ith data point is from the jth gaussian distribution, indicating the probability that the data point belongs to the distribution. Alpha, beta, delta are coefficients used to adjust the complexity of the model and the influence of the data features, allowing the model to more flexibly adapt to different characteristics of the data, alpha is determined by evaluating the influence of different mixing coefficients pi _j on the model performance, beta is determined by considering the complexity of the covariance matrix sigma _j on the model influence, and delta is determined by evaluating the influence of the distance between the data point x _i and each gaussian distribution mean mu _j on the classification result. Pi _j is the mixing coefficient of the j-th distribution, and represents the proportion of the gaussian distribution in the whole distribution. Sigma _j is the covariance matrix of the jth Gaussian distribution, describing the correlation between the variables in the distribution. x _i is a data point representing observed borrower credit data. Mu _j is the mean value of the j-th Gaussian distribution and represents the location of the center point of the distribution. k is the number of distributions, representing the total number of gaussian distributions assumed in the model. n is the number of data points, indicating how many borrower's credit data in total are analyzed. L (x _i, C) is an additional term about data point x _i and its confidence level C, introducing additional information into the model, improving the accuracy of the prediction. λ is an additional term coefficient that adjusts the intensity of the impact of the credit rating information on the final model, as determined by measuring the contribution of the additional term L (x _i, C) to model accuracy.

The bayesian information criterion is according to formula II:

BIC＝-2ln(L)+kln(n)

And determining the optimal distribution quantity and generating a behavior pattern classification result.

Where L is the maximum value of the likelihood function of the model, i.e. the likelihood function value at which the model parameters are optimized for a given data. The degree of fitting of the model to the observed data is reflected, and the larger the likelihood function is, the higher the probability of the data under the current model and parameters is. k is the number of parameters in the model. Including all parameters that need to be estimated, such as the mean, covariance matrix of each gaussian distribution in a gaussian mixture model, and the mixture coefficients all need to be estimated, all taking into account k. n is the number of data points, i.e., the sample size. Representing the total number of all borrower records. ln represents a natural logarithmic function. BIC penalizes likelihood function values of a model, and penalty terms depend on the number of parameters and the number of data points in the model, so as to avoid overfitting. The more parameters in the model, or the larger the amount of data, the larger the penalty term for the BIC.

The execution process is as follows:

Credit data for the borrower is collected, including repayment actions, account information, and the like. The data is preprocessed, such as cleaning, standardization, etc., to ensure the quality of the input data.

An initial distribution number k is selected. For example, different k values may be tried gradually from small to large. Preliminary estimates of the parameters (mean μ _j, covariance matrix Σ _j) and mixing coefficient pi _j for each gaussian distribution are made.

E-step (desired step) by using the current model parameters, the probability (posterior probability) γ (z _ij)₀) that each data point belongs to each Gaussian distribution is calculated.

M-step (maximization step) maximizes the desired log-likelihood function calculated by E-step by updating the model parameters, including the mean μ _j, covariance matrix Σ _j, and mixing coefficient pi _j for each gaussian distribution.

The parameters in equation I are added to E-step and M-step, including the weight coefficient w _ij, the adjustment coefficients α, β, δ, and the additional term L (x _i, C) for the credit rating and its coefficient λ.

For each k value, the BIC value is calculated according to equation II using the modified model and parameters. Where L is the maximum 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.

For different k values, their corresponding BIC values are compared. K with the smallest BIC value is chosen as the optimal number of distributions, since smaller BIC values indicate a better fit of the model to the data while taking the model complexity into account.

Classification of the borrower's credit patterns is accomplished by assigning data points to the most likely gaussian distribution using the optimal k values and corresponding model parameters.

S102: based on the behavior pattern classification result, analyzing repayment frequency and loan amount usage, and identifying a plurality of credit level characteristics including payment behavior, account age and credit utilization rate, wherein the specific process of generating a credit level difference analysis result is as follows;

S102, based on a behavior pattern classification result, adopting logistic regression analysis, using a Logit function of statsmodels library, extracting characteristics such as repayment frequency, loan amount use and the like, taking the characteristics as independent variables, setting up dummy variables to process classification data such as payment behavior, account age and credit utilization rate, setting C (payment behavior) +C (account age) +credit utilization rate as a model formula, calling a fit method to perform model training, using a predict method to predict credit level difference based on the trained model, setting an output threshold to be 0.5 to distinguish high credit level from low credit level, and generating a credit level difference analysis result.

S103: the credit level difference analysis result is utilized to adjust the classification standard, optimize the credit level group division and generate a credit grading model of a borrower;

S103, based on the credit level difference analysis result, adopting a K-means clustering algorithm, using KMeans functions of scikit-learn libraries, designating n_ clusters as the number of groups determined through credit level difference analysis, selecting K-means++ to optimize the selection of initial clustering centers by init parameters, setting max_iter to be 300 so as to ensure algorithm convergence, performing 10 independent operations for n_init to select the best result, calling a fit_ predict method to cluster credit data, and dynamically adjusting classification standards by calculating the average distance from data points in each group to the clustering center of the data points to generate a credit person credit grading model.

As shown in fig. 3, the steps of tracking the borrower credit activity data, capturing the credit activity change and recording the key event based on the borrower credit rating model, and generating the borrower credit dynamic deviation record further comprise:

S201: monitoring activities of a daily credit account of a borrower by adopting a borrower credit grading model, wherein the activities comprise current loan application and repayment behaviors, and the specific process for generating a daily credit activity record is as follows;

The S201 substep is based on a credit rating model of a borrower, AND uses SQL query sentences to retrieve daily credit account activities of the borrower from a database, wherein the SQL query sentences comprise a SELECT sentence for a loan application form AND a repayment record form, a WHERE clause is set to be in a date range of the current day, AND conditions screen records with account states active, JOIN operations connect a user basic information form to acquire complete identity information of the borrower, ORDER BY clauses ORDER the results in time sequence, a pandas library of Python is utilized to convert data frames of the query results, each record is marked with a timestamp AND an activity type, the current loan application state AND repayment behavior are included, AND a daily credit activity record is generated.

S202: screening abnormal activities from daily credit activity records, wherein the abnormal activities comprise credit score decrease or newly added overdue records, and marking the abnormal activities as key credit change events, and the specific process of generating key credit event records is as follows;

S202, the substep is based on daily credit activity records, abnormal detection is achieved by adopting a Python script, the sensitivity of credit score reduction and the identification rule of overdue records are set through defining threshold variables, if-else logic is used for judging whether each record meets the condition of abnormal activity, and for the condition that the credit score reduction or the newly added overdue records are met, an application method is called to add the records into a new list, a unique identifier is allocated for each abnormal activity, and the occurrence time is marked by using datetime libraries to generate key credit event records.

S203: analyzing the key credit event record, sorting the variation trend and key deviation point of credit activity, evaluating the health condition of credit activity, and generating the dynamic deviation record of credit of borrower as follows;

The S203 substep is based on the key credit event record, a data visualization tool such as Matplotlib and Seaborn library is adopted to draw a credit activity change trend chart, a chart type is set to be a time sequence chart, an X-axis represents time, a Y-axis represents key credit activity indexes including credit score and overdue record quantity, a change trend of each activity is drawn through a plot function, key deviation points are highlighted through a plot function, a describe method is called to carry out statistical analysis on the key events including average value, median and standard deviation, and the credit activity change trend and the key deviation points are arranged through comparison of the statistical data of the key indexes and the chart analysis, so that a borrower credit dynamic deviation record is generated.

As shown in fig. 4, the step of analyzing the behavior preference data of the borrower under the economic pressure and the behavior trend causing overdue by the borrower credit dynamic deviation record to generate the overdue behavior dynamic analysis result further comprises:

S301: the method comprises the specific processes of collecting credit score variation, newly-added loan application times and current overdue frequency data through dynamic deviation records of the borrower credit, screening abnormal credit behaviors of the borrower under economic pressure, and obtaining a credit activity overview of the borrower under economic pressure;

S301, the substep is based on the dynamic deviation record of the credit of the borrower, data collection is executed, the credit score change, the newly added number of loan application and the extraction of the current overdue frequency data are included, a data mining technology is adopted, required fields are extracted from a database through SQL query, data cleaning is carried out by using a Pandas library of Python, the records including missing value processing, data type conversion, marked reduction of the credit score, sharp increase of the loan application and rising of the overdue frequency are used as abnormal credit behaviors, data points falling in an abnormal range are screened through setting a threshold value, the change trend of the credit score and the overdue frequency is further analyzed through drawing a time sequence diagram, and the credit activity overview of the borrower under economic pressure is obtained.

S302: based on the credit activity overview of the borrower under the economic pressure, analyzing a plurality of change time points of abnormal expenditure increase or credit card use frequency increase, and identifying and marking behavior trends causing overdue through the change time points, wherein the specific process of obtaining key behavior change records is as follows;

S302, based on the credit activity overview of the borrower under economic pressure, executing the consumption mode analysis of the borrower, modeling consumption data by utilizing an ARIMA model in statsmodels library of Python by adopting a time sequence analysis method, setting parameters to automatically identify the optimal differential order d, autoregressive item p and moving average item q, determining a parameter range in an auxiliary way through ACF and PACF graphs, identifying time points when abnormal expenditure increases or credit card use frequency increases, comparing the model prediction result with actual data, identifying behavior trend causing overdue through difference analysis, and obtaining a key behavior change record.

S303: the critical behavior change record is adopted to evaluate overdue possibility of a borrower under various economic pressures, and the relation between behavior preference and overdue risk is distinguished, so that the specific process of obtaining the overdue behavior dynamic analysis result is as follows;

The S303 substep adopts key behavior change record, performs overdue possibility assessment, adopts a machine learning classification algorithm, implements a random forest algorithm through a scikit-learn library of Python, takes the consumption mode and credit behaviors of borrowers as characteristic input, takes overdue or not as target variables, sets n_ estimators as 100 to construct a forest of 100 trees, sets max_depth as None allowed tree depth growth until all leaves are pure or leaves containing less than min_samples_split samples, uses a train_test_split function to divide the data into a training set and a testing set, performs model training on training data by a fit method, predicts overdue conditions of the testing set by a predict method, further analyzes matching degree of an evaluation result and actual overdue record, and performs characteristic importance analysis to distinguish relation between behavior preference and overdue risks, so as to obtain overdue behavior dynamic analysis results.

As shown in fig. 5, the step of using the result of the overdue behavior dynamic analysis to construct a prediction model to predict the overdue probability of the borrower and obtain the result of the borrower overdue risk analysis further includes:

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

S401 substep is based on the result of the overdue action dynamic analysis, the data are sorted by using Python language in combination with Pandas library, credit activity deviation and overdue action trend under DATAFRAME object aggregate economic pressure are grouped by month by applying groupby method to each borrower, average credit score change and overdue event count of each group are calculated by using agg function, economic index data are combined by merge method, and overdue action and credit activity comprehensive data set is generated.

S402: analyzing the behavior data of the borrower based on the overdue behavior and credit activity comprehensive data set by combining an accumulated prospect theory analysis method, and identifying a behavior mode associated with overdue probability, wherein the specific process of constructing the overdue probability prediction model is as follows;

s402, based on the overdue behavior and credit activity comprehensive data set, analyzing by adopting an accumulated prospect theory analysis method, constructing a logistic regression model by using a Python language and combining with a Scikit-learn library, setting a software parameter as liblinear to adapt to the scale and the characteristics of the data set by using a LogisticRegression type initialization model, inputting behavior data and overdue labels by using a fit method, performing model training, identifying the association between overdue probability and behavior modes, and constructing an overdue probability prediction model.

The cumulative prospect theory analysis method is according to an improved formula III:

Calculating behavior data of a borrower to obtain overdue probability values of the borrower, and constructing a overdue probability prediction model according to the overdue probability values;

Where V '(B) represents the improved cumulative prospect value reflecting the anticipated value of the borrower's overdue risk under economic stress. i represents an index of actions for traversing all possible actions or events. gains refers to the set of actions that all lead to revenue. losses refers to the set of actions that all lead to loss. Pi' (p _i, θ) represents an improved weighting function, p _i is the occurrence probability of behavior i, θ is a newly introduced parameter, reflects the influence of market fluctuation rate or credit market dynamics on probability perception, and is determined by market fluctuation index or credit market overdue rate historical data. v ' (x _i, α, β, λ) represents an improved cost function, where x _i is the credit score change or economic impairment resulting from behavior i, α and β are parameters for adjusting the gain and loss sensitivity, respectively, determined by analyzing the credit behavior sensitivity of the borrower under different economic conditions, and by analysis of the borrower's reaction strength to loss, λ is a dynamic adjustment parameter for the loss aversion coefficient, determined dynamically by evaluating the borrower's loss aversion behavior, and combining with market trend data. x _i represents the credit score change or economic benefit resulting from action i. p _i represents the probability of occurrence of behavior i.

The execution process is as follows:

Credit activity data for the borrower is collected, including credit activity bias and overdue behavioral trends under economic pressure, as well as market volatility or dynamic data for the credit market.

And determining a parameter theta in the weighting function by using market fluctuation index or overdue rate historical data, and reflecting the influence of market dynamics on probability perception.

The parameters α, β, and λ in the cost function are dynamically adjusted based on borrower historical behavior data and market conditions. The parameters respectively adjust the gain sensitivity, the loss sensitivity and the loss aversion degree, and accurately reflect the behavior modes of borrowers under different economic conditions.

For each behavior or event i, a modified weighting function pi' (p _i, θ) is calculated, and the original occurrence probability p _i is adjusted by the parameter θ, taking into account market fluctuations or the effects of other external factors.

For each behavior or event i, a modified cost function v' (x _i, α, β, λ) is calculated. The credit score change or economic profit-and-loss x _i is caused according to the behavior. And parameters α, β, λ calculate value reflecting borrower sensitivity to gain and loss aversion emotion.

And accumulating the products of the improved cost function and the weighting function under all profit conditions through a formula III, accumulating the corresponding products under all loss conditions, and finally adding the two results to obtain an improved accumulated prospect value V' (B).

The improved cumulative prospect value V '(B) is used to predict the borrower's overdue probability. In the model, V' (B) can be used to evaluate overdue risk in different behavioral patterns, thereby providing support for risk management and decision making.

S403: calculating the overdue probability of the borrower by comprehensively referring to credit activity deviation under economic pressure by using the overdue probability prediction model, wherein the specific process for obtaining the overdue risk analysis result of the borrower is as follows;

S403, the substep utilizes a overdue probability prediction model, comprehensively uses Python language and Numpy library to conduct numerical processing on credit activity deviation under economic pressure, calculates overdue probability of a borrower based on a trained logistic regression model through a predict-proba method, and utilizes argmax function to select the highest overdue risk level from probability results for each borrower to generate a borrower overdue risk analysis result.

As shown in fig. 6, the steps of classifying the loan risk according to the result of the analysis of the risk of overdue, analyzing the key factors causing the variation of the overdue probability, and generating the risk classification and the recognition result of the overdue factors further comprise:

s501: the method comprises the specific processes of dividing the risk level of a borrower according to the overdue risk analysis result of the borrower and referring to the overdue probability value to obtain a risk level list;

The S501 substep executes risk classification based on the analysis result of the overdue risk of the borrower, processes data through Pandas library by using a Python script by adopting a threshold segmentation method, sets the numerical range of the overdue probability as a threshold, classifies the borrower with the overdue probability lower than 5% as low risk, classifies the borrower with the overdue probability between 5% and 20% as medium risk, classifies the borrower with the overdue probability higher than 20% as high risk, maps the overdue probability numerical value to the corresponding risk level by using a cut function, generates DATAFRAME containing the borrower ID and the corresponding risk level, and generates a risk level list.

S502: analyzing common characteristics and behaviors of each risk level borrower based on the risk level list, and identifying a plurality of key factors affecting overdue probability changes including income level, employment stability and liability ratio, wherein the specific process for generating the key overdue factor list is as follows;

S502, performing feature and behavior analysis based on a risk level list, performing generalized linear model analysis through a glm function of R language by adopting a multivariate analysis method, taking income level, employment stability and liability ratio as independent variables, setting family parameters as binomial for logistic regression, identifying common features and behaviors of groups of borrowers with different risk levels, performing variance analysis on the model through anova functions, checking influence of the respective variables, identifying factors which have obvious influence on overdue probability change, and generating a key overdue factor list.

S503: carrying out iterative analysis on the borrower groups of each risk level by utilizing the key overdue factor list, revealing the relevance among each type of factors, and obtaining the specific processes of risk classification and overdue factor identification result;

s503 substep uses a key overdue factor list to execute iterative analysis, adopts a clustering analysis method, sets n_ clusters parameters as the number of different risk grades in a borrower group through KMeans algorithm in scikit-learn library of Python, carries out subdivision inside the borrower group, carries out clustering analysis on the borrower group of each risk grade according to income level, employment stability and liability ratio by the fit_ predict method, reveals the interrelation among different key overdue factors and the difference of overdue risk influence thereof inside the same risk grade, and obtains risk classification and overdue factor identification results.

As shown in fig. 7, the step of analyzing the risk classification and the correlation of the overdue factors in the overdue factor recognition result to identify the interaction between each factor and its influence on the overdue risk, and further includes:

s601: the method comprises the steps of performing data arrangement on overdue factors in risk classification and overdue factor identification results, collecting overdue probability, economic factors and multiple key data of personal behavior characteristics, and performing classification integration, wherein the specific process of generating overdue factor data sets is as follows;

S601, based on risk classification and overdue factor recognition results, executing data arrangement, adopting a data aggregation technology, operating the data through Pandas libraries by using a Python script, grouping the data according to risk levels by using groupby functions, calculating average values of overdue probability under each risk level, economic factors such as median of income level and average values of personal behavior characteristics such as liability ratio by using agg functions, carrying out data classification integration, merging the results into one DATAFRAME, providing a basis for subsequent analysis, generating overdue factor data sets, and generating overdue factor data sets.

S602: matching and comparing multiple data of the overdue factor data set to reveal the interaction among each factor of income level, employment stability and liability ratio, and the specific process of generating the overdue factor correlation analysis table is as follows;

The S602 substep is based on the overdue factor data set, executes the correlation analysis among factors, adopts a Pearson correlation coefficient calculation method, compares factors such as income level, employment stability, liability ratio and the like by using corrcoef functions through a NumPy library of Python, calculates correlation coefficients, draws a heat map through a matplotlib library to visualize the correlation among the factors, reveals how interactions between different economic factors and personal behavior characteristics affect overdue probability, generates an overdue factor correlation analysis table, and generates an overdue factor correlation analysis table.

S603: based on the overdue factor correlation analysis table, carrying out correlation scoring, and quantifying influence intensity and action mode among evaluation factors to obtain a specific process of risk factor correlation analysis result;

S603, executing a relevance grade based on a overdue factor relevance analysis table, modeling the linear relation between overdue probability and each factor by using an lm function through an R language, setting the overdue probability as a dependent variable, setting a income level, employment stability and liability ratio as independent variables, acquiring a coefficient estimated value and a p value of a model through a summary function, quantitatively evaluating the influence intensity and an action mode of each factor on the overdue probability, carrying out diagnosis and inspection on the model to ensure that no multiple co-linearity problem exists, obtaining the relevance grade of each factor on overdue risk influence, and generating a risk factor relevance analysis result.

Although the application has been described in connection with specific features and embodiments thereof, it will be apparent that various modifications and combinations can be made without departing from the spirit and scope of the application. Accordingly, the specification and drawings are merely exemplary illustrations of the present application as defined in the appended claims and are considered to cover any and all modifications, variations, combinations, or equivalents that fall within the scope of the application. It will be apparent to those skilled in the art that various modifications and variations can be made to the present application without departing from the scope of the application. Thus, it is intended that the present application cover the modifications and variations of this application provided they come within the scope of the appended claims and their equivalents.

Claims (8)

1. An intelligent identification and analysis method for financial overdue risk is characterized by comprising the following steps:

Identifying groups of borrowers with multiple credit classes by carrying out cluster analysis on the credit data of the borrowers, determining the cluster number by using the evaluation index, and generating a credit grading model of the borrowers;

Tracking the credit activity data of the borrower based on the credit grading model, capturing the credit activity change and recording key events to generate a credit dynamic deviation record of the borrower;

Analyzing behavior preference data of the borrower under economic pressure and behavior trend causing overdue through the credit dynamic deviation record of the borrower, and generating overdue behavior dynamic analysis results;

Constructing a prediction model by using the overdue behavior dynamic analysis result, predicting overdue probability of the borrower, and obtaining overdue risk analysis result of the borrower;

Dividing the risk level of the borrower according to the overdue risk analysis result of the borrower, and analyzing key factors causing overdue probability change to generate risk classification and overdue factor identification results;

And analyzing the association of the overdue factors in the risk classification and overdue factor identification results, and identifying the interaction among each factor and the influence of each factor on overdue risk to obtain a risk factor association degree analysis result.

2. The intelligent identification and analysis method for financial overdue risk according to claim 1, wherein: the borrower credit rating model comprises a credit rating interval, a overdue record classification and a financial health index, the borrower credit dynamic deviation record comprises credit score decreasing events, new loan application times and overdue occurrence frequency, the overdue behavior dynamic analysis result comprises consumption mode change, abnormal expenditure increase and credit card use frequency under economic pressure, the borrower overdue risk analysis result comprises overdue probability numerical values, overdue occurrence time period prediction and overdue risk level division of each borrower, the risk classification and overdue factor identification result comprises a borrower group classified according to the overdue probability, overdue economic factors and personal behavior characteristics, and the risk factor association degree analysis result comprises association strength of financial conditions and personal behavior characteristics, interaction modes among multiple economic factors and compound factor analysis for influencing overdue risks.

3. The method of claim 1, wherein the step of identifying a plurality of classes of credit groups by performing a cluster analysis on the credit data, determining a number of clusters using the evaluation index, and generating a credit rating model further comprises:

The credit data of the borrower is subjected to clustering analysis, the repayment behaviors and account information are monitored through a Gaussian mixture model by adopting an expectation maximization algorithm, a credit behavior mode is identified, the distribution quantity is determined through a Bayesian information quantity criterion, and a behavior mode classification result is generated;

based on the behavior pattern classification result, analyzing repayment frequency and loan amount usage, identifying a plurality of credit level characteristics including payment behavior, account age and credit utilization rate, and generating a credit level difference analysis result;

And adjusting the classification standard by utilizing the credit level difference analysis result, optimizing credit level group division, and generating a borrower credit grading model.

4. The intelligent identification and analysis method of financial overdue risk according to claim 3, wherein the gaussian mixture model and the expectation maximization algorithm are according to the improved formula I:

subdividing and identifying financial overdue risks by referring to the borrower characteristics and the credit rating information;

Where Q' (θ|θ ^(t)) is the log likelihood function expected for the improved condition given the current parameter estimate θ ^(t), w _ij represents the weight coefficient of the ith data point at the jth Gaussian distribution, γ (z _ij) is the posterior probability of the ith data point from the jth Gaussian distribution, α, β, δ is the coefficient for adjusting the model complexity and the data characteristic impact, pi _j is the mixing coefficient of the jth distribution, Σ _j is the covariance matrix of the jth Gaussian distribution, x _i is the 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 an additional term for data point x _i and its confidence level C, λ is the additional term coefficient;

The bayesian information criterion is according to formula II:

BIC＝-2ln(L)+kln(n)

determining the optimal distribution quantity and generating a behavior pattern classification result;

Where L is the maximum of the likelihood function of the model, k is the number of parameters in the model, n is the number of data points, ln is the natural logarithmic function, and BIC penalizes the likelihood function value of the model.

5. The method of claim 1, wherein tracking borrower credit activity data based on the borrower credit rating model, capturing credit activity changes and recording key events, generating a borrower credit dynamic deviation record, further comprising:

Monitoring activities of a daily credit account of a borrower by adopting the borrower credit rating model, wherein the activities comprise current loan application and repayment behaviors, and generating a daily credit activity record;

screening abnormal activities from the daily credit activity records, wherein the abnormal activities comprise credit score decrease or newly added overdue records, marking the abnormal activities as key credit change events, and generating key credit event records;

And analyzing the key credit event records, sorting the variation trend and key deviation points of credit activities, evaluating the health condition of the credit activities, and generating a dynamic credit deviation record of a borrower.

6. The method of claim 1, wherein the step of analyzing the behavior preference data of the borrower under economic pressure and the behavior trend causing overdue by the borrower credit dynamic deviation record to generate overdue behavior dynamic analysis results further comprises:

collecting credit score variation, newly-added loan application times and current overdue frequency data through the dynamic deviation record of the borrower credit, screening abnormal credit behaviors of the borrower under economic pressure, and obtaining a credit activity overview of the borrower under economic pressure;

Based on the credit activity overview of the borrower under the economic pressure, analyzing a plurality of change time points of abnormal expenditure increase or credit card use frequency increase, and identifying and marking behavior trends causing overdue through the change time points to obtain key behavior change records;

And evaluating overdue possibility of the borrower under various economic pressures by adopting the key behavior change record, distinguishing the relation between behavior preference and overdue risk, and obtaining the overdue behavior dynamic analysis result.

7. The method of claim 1, wherein the step of constructing a predictive model using the results of the dynamic analysis of overdue actions to predict the overdue probability of a borrower and obtain the results of the analysis of overdue risk of the borrower further comprises:

Generating a comprehensive data set of overdue behaviors and credit activities by arranging the overdue behavior dynamic analysis results, including credit activity deviation and overdue behavior trend under economic pressure;

analyzing the behavior data of the borrower based on the overdue behavior and credit activity comprehensive data set by combining an accumulated prospect theory analysis method, identifying a behavior mode associated with overdue probability, and constructing an overdue probability prediction model;

And calculating the overdue probability of the borrower by comprehensively referring to the credit activity deviation under the economic pressure by using the overdue probability prediction model, and obtaining the overdue risk analysis result of the borrower.

8. The intelligent identification and analysis method of financial overdue risk according to claim 7, wherein the cumulative prospect theory analysis method is according to improved formula III:

Calculating behavior data of a borrower to obtain overdue probability values of the borrower, and constructing a overdue probability prediction model according to the overdue probability values;

Where V ' (B) represents the modified cumulative prospect value, i represents the index of behavior, gains represents the set of behavior that all results in revenue, losses represents the set of behavior that all results in loss, pi ' (p _i, θ) represents the modified weighting function, p _i is the probability of occurrence of behavior i, θ reflects the impact of market volatility or credit market dynamics on probability perception, V ' (x _i, α, β, λ) represents the modified cost function, x _i represents the credit score change or economic loss caused by behavior i, and p _i represents the probability of occurrence of behavior i.