Credit variable-based credit label determination method and system by using AI

Abstract

The invention relates to the technical field of credit label determination, in particular to a credit variable-based credit label determination method and system by using AI. The method comprises the following steps: firstly, collecting and sorting multi-dimensional data of a customer, and constructing a feature vector capable of comprehensively representing the credit status of the customer; secondly, calculating influence weights of different credit characteristics on credit evaluation by using a machine learning algorithm and learning existing samples; then, the clients are layered based on the feature weight coefficient and the credit score interval, and a credit fluctuation coefficient is introduced to quantify the deviation degree between the actual credit performance of the clients and the expected level; and finally, factors such as the feature weight, the credit fluctuation coefficient and sample distribution of various clients are incorporated into a comprehensive scoring model, and a final credit label of the client is scientifically determined. Through multi-dimensional data analysis and credit fluctuation coefficient design, more accurate credit evaluation is realized, and meanwhile, the modeling cost and difficulty are remarkably reduced.

Description

Technical Field

[0001] This application relates to the technical field of credit label determination, and in particular, to a method and system for determining credit labels based on credit variables using AI. Background Art

[0002] With the rapid development of fintech, credit risk assessment plays an increasingly important role in the operation and management of financial institutions. Accurate credit risk assessment not only relates to the asset quality of financial institutions but also is an important foundation for the development of inclusive finance. Therefore, establishing a scientific and effective credit assessment system has become a key task in the financial industry.

[0003] The prior art mainly uses an automated scoring system combined with multi-dimensional data analysis for credit risk assessment. Such systems collect customers' transaction records, behavioral data, and external information to establish a scoring model to predict customers' credit risk levels and make credit-granting decisions accordingly.

[0004] However, in the prior art, only customers who have completed the loan cycle can obtain credit labels, and a large number of potential customers lack complete loan performance, resulting in an insufficient number of samples with credit labels. This sample imbalance problem reduces the accuracy of the model, increases the cost and difficulty of credit risk assessment, and this situation needs to be further improved. Summary of the Invention

[0005] To solve the problems of high cost and great difficulty in credit risk assessment of the existing automated scoring system, this application provides a method and system for determining credit labels based on credit variables using AI, and adopts the following technical solutions: In the first aspect, this application provides a method for determining credit labels based on credit variables using AI, including the following steps: Obtain a credit feature vector representing the customer's credit status according to multi-dimensional data including customer information; Based on the credit feature vector, use a machine learning algorithm to calculate feature weight coefficients representing the influence of different credit features on credit assessment; According to the feature weight coefficients and credit score intervals, stratify and classify customers and calculate the credit fluctuation coefficients of each type of customer, where the credit fluctuation coefficients represent the degree of deviation between the customer's actual credit performance and the expected credit level; Based on the feature weight coefficients, credit fluctuation coefficients, and the sample sizes of various types of customers, determine the final credit label of the customer through a comprehensive scoring model, where the credit label reflects the overall credit level and risk degree of the customer.

[0006] By adopting the above technical solution, this application first collects and organizes multi-dimensional data including customer basic information, transaction behaviors, social relationships, etc., and constructs a feature vector that can comprehensively represent the customer's credit status. Secondly, machine learning algorithms are used to calculate the influence weights of different credit features on credit assessment through learning from existing samples. Then, based on the feature weight coefficients and credit score intervals, customers are stratified, and a credit fluctuation coefficient is introduced to quantify the deviation degree of the customer's actual credit performance from the expected level. Finally, factors such as feature weights, credit fluctuation coefficients, and the sample distributions of various types of customers are incorporated into the comprehensive scoring model to scientifically determine the final credit label of the customer. Through multi-dimensional data analysis and the design of credit fluctuation coefficients, more accurate credit assessment is achieved, and at the same time, the modeling cost and difficulty are significantly reduced.

[0007] Optionally, the credit feature vector includes transaction behavior features, financial features, credit record features, and association features. Obtaining a credit feature vector that represents the customer's credit status based on multi-dimensional data including customer information specifically includes the following steps: Obtain the customer's historical transaction data and extract transaction behavior features including transaction frequency, transaction scale, and transaction time distribution. Obtain the customer's financial statement data and extract financial features including asset scale, liability level, and cash flow status. Obtain the customer's credit investigation data and extract credit record features including historical default records, overdue situations, and credit card usage status. Obtain the customer's related party data and extract association features including the operating conditions of related enterprises, equity relationships, and industry credit levels. Perform feature fusion on the transaction behavior features, financial features, credit record features, and association features to obtain a credit feature vector that represents the customer's credit status.

[0008] By adopting the above technical solution, this application first introduces the transaction behavior dimension. By analyzing the customer's transaction frequency, transaction scale, and time distribution characteristics, the business activity and stability of the customer are revealed. Secondly, in-depth analysis of the financial statement data is carried out, and key indicators such as asset scale, liability level, and cash flow status are extracted to evaluate the financial health of the customer. Then, the credit investigation data is integrated, including historical default records, overdue situations, and credit card usage status, to comprehensively reflect the customer's credit performance ability. Finally, the association dimension analysis is introduced. By examining the operating conditions of related enterprises, the equity relationship network, and the industry credit level, potential transmission risks are warned. Finally, feature fusion technology is adopted to scientifically combine the features of the above four dimensions to construct a comprehensive and three-dimensional credit feature vector. Through the organic integration of multi-dimensional data, the comprehensiveness and accuracy of credit assessment are significantly improved, and the risks brought by single-dimensional assessment are effectively avoided.

[0009] Optionally, based on the credit feature vector, a feature weight coefficient representing the influence magnitude of different credit features on credit assessment is calculated using a machine learning algorithm, which specifically includes the following steps: Rank the importance of the transaction behavior features and calculate the transaction behavior time decay coefficient; Conduct industry grouping analysis on the financial features and calculate the industry deviation coefficient; Conduct default probability analysis on the credit record features and calculate the default impact coefficient; Conduct association conduction analysis on the association features and calculate the association conduction coefficient; Based on the time decay coefficient, industry deviation coefficient, default impact coefficient, and association conduction coefficient, obtain the feature weight coefficients of each credit feature.

[0010] By adopting the above technical solution, the present application first introduces time dimension analysis. By ranking the importance of transaction behavior features and calculating the time decay coefficient, accurate characterization of recent behaviors is achieved; secondly, an industry grouping analysis method is adopted. By calculating the deviation degree between financial features and industry standards, the deviation coefficient reflecting industry characteristics is obtained; then, default probability analysis is carried out. Through statistical modeling of historical credit records, the default impact coefficient is calculated to quantify the degree of credit risk; finally, an association conduction analysis framework is constructed. By evaluating the risk conduction effect between related parties, the association conduction coefficient is calculated to warn of potential risks; finally, based on the coefficients of the above four dimensions, integration is carried out to obtain dynamically adjusted feature weight coefficients; multi-dimensional dynamic weight calculation significantly improves the accuracy and timeliness of feature weights and effectively solves the limitations of traditional static weight methods.

[0011] Optionally, according to the feature weight coefficient and credit score interval, customers are stratified and classified, and the credit fluctuation coefficient of each type of customer is calculated, which specifically includes the following steps: Combine the feature weight coefficient and credit score interval to classify the customer sample and divide the customers into different credit grade categories; Calculate the historical credit score sequence of customers within each credit grade category to obtain the credit score fluctuation range within the category; Analyze the fluctuation trajectory of each customer within the score interval, calculate the deviation degree between the fluctuation trajectory and the category fluctuation range, and obtain the credit stability index of the customer individual; Statistically analyze the distribution of stability indicators of customers within each credit grade category to determine the overall fluctuation characteristics of the category; According to the stability index of the customer individual and the category fluctuation characteristics, calculate the credit fluctuation coefficient of each type of customer.

[0012] Since traditional credit assessment methods often only focus on the credit status of customers at a certain point in time and ignore the dynamic change characteristics of credit performance, it is impossible to effectively identify high-risk customers with large fluctuations in credit levels. By adopting the above technical solutions, this application first stratifies customers based on feature weight coefficients and credit score intervals, dividing customers into different credit rating categories. Secondly, by calculating the historical credit score sequences of customers within each category, a credit score fluctuation range unique to the category is obtained, forming a dynamic reference standard. Thirdly, by calculating the deviation degree between the individual score trajectory and the category fluctuation range, the credit stability of customers is quantified. Further, statistical analysis is performed on the distribution of stability indicators within each credit rating category to determine the overall fluctuation characteristics of the category and establish a group benchmark. Finally, by comprehensively considering individual stability indicators and category fluctuation characteristics, a credit fluctuation coefficient that can accurately reflect credit fluctuation risk is calculated. Through the hierarchical dynamic analysis method, a comprehensive assessment of customer credit stability is achieved, effectively enhancing the early warning ability of credit risk.

[0013] Optionally, analyze the fluctuation trajectory of each customer within the score interval and calculate the deviation degree between the fluctuation trajectory and the category fluctuation range, which specifically includes the following steps: Obtain the continuous score data of the customer during the evaluation period and calculate the maximum value, minimum value, and mean value of the score sequence. Set a fluctuation tolerance parameter according to the credit level of the credit rating category. The higher the credit level, the smaller the fluctuation tolerance parameter; the lower the credit level, the larger the fluctuation tolerance parameter. Based on the category fluctuation tolerance parameter and historical fluctuation data, calculate the upper and lower limits of the fluctuation benchmark for each credit rating category. Calculate the overlap degree and exceedance degree between the fluctuation interval of the customer score sequence and the corresponding category fluctuation benchmark interval respectively to obtain the deviation degree considering the fluctuation tolerance.

[0014] Traditional volatility analysis methods often adopt a unified volatility standard, failing to consider the differences in volatility sensitivity among customers with different credit levels. As a result, the small volatility risks of high-credit-level customers are underestimated, while the normal volatility of low-credit-level customers is over-amplified. By adopting the above technical solution, this application first collects the continuous scoring data of customers during the evaluation period, calculates the maximum value, minimum value, and mean value of the scoring sequence, and establishes a basic evaluation framework. Secondly, a volatility tolerance parameter mechanism is introduced, which is dynamically adjusted according to the credit levels of different credit rating categories. A more stringent volatility tolerance standard is set for high-credit-level customers, and a larger volatility allowance space is given to low-credit-level customers. Thirdly, based on the set volatility tolerance parameters and historical volatility data, the upper and lower limits of the volatility benchmark unique to each credit rating category are scientifically calculated to construct a differentiated evaluation interval. Finally, by calculating the overlap and exceedance degrees between the volatility interval of the customer scoring sequence and the category volatility benchmark interval, a more targeted deviation degree evaluation result is obtained, realizing precise volatility monitoring of customers with different credit levels and effectively improving the accuracy of risk warning.

[0015] Optionally, based on the characteristic weight coefficient, credit volatility coefficient, and the sample quantity of each type of customer, the final credit label of the customer is determined through a comprehensive scoring model, which specifically includes the following steps: Multiply the credit characteristics of the customer by the corresponding characteristic weight coefficient to obtain the weighted characteristic score. According to the credit volatility coefficient of the category to which the customer belongs, correct the weighted characteristic score to obtain an adjusted score considering volatility. Based on the sample quantity distribution of each type of customer, calculate the sample sparsity coefficient. Statistically analyze the score distribution characteristics of each credit rating category to determine the classification threshold of the credit label. Determine the final credit label of the customer according to the adjusted score, sample sparsity coefficient, and classification threshold of the customer.

[0016] By adopting the above technical solution, this application first performs weighted calculation on the credit characteristics of the customer through the characteristic weight coefficient to obtain the basic characteristic score. Secondly, a credit volatility coefficient is introduced for score correction, and by quantitatively analyzing the volatility characteristics of the category to which the customer belongs, dynamic adjustment of the score is realized. Then, considering the influence of the sample distribution, by calculating the sample sparsity coefficient, the influence degree of the sample quantity on the determination result is scientifically evaluated. Further, by analyzing the score distribution characteristics of each credit rating category, a dynamic classification threshold system is established. Finally, by comprehensively considering the adjusted score, sample sparsity, and classification threshold, a more representative final credit label is obtained. Through the organic combination of multi-dimensional factors, the determination deviation problem of traditional methods in the case of sample imbalance is effectively solved, and the accuracy and reliability of credit label determination are significantly improved.

[0017] Optionally, the adjustment score includes a static adjustment score and a dynamic adjustment score. The acquisition of the adjustment score specifically includes the following steps: Divide the historical data of the customer into a long-term evaluation window and a short-term evaluation window according to the time dimension, which are respectively used to calculate the static credit level and the dynamic credit trend; Calculate the long-term credit stability coefficient and the short-term credit fluctuation coefficient respectively according to the historical evaluation data; Obtain the static adjustment score and the dynamic adjustment score of the customer respectively according to the long-term credit stability coefficient and the short-term credit fluctuation coefficient.

[0018] Traditional adjustment score calculation methods often only focus on the credit performance in a single time dimension, and fail to effectively balance the long-term stability and the short-term change trend, resulting in the lack of timeliness and continuity in the evaluation of the customer's credit level; by adopting the above technical solution, the present application first introduces a time dimension stratification, divides the customer historical data into a long-term evaluation window and a short-term evaluation window, which are respectively used to evaluate the customer's basic credit level and the latest development trend; by analyzing the long-term data, calculate the credit stability coefficient to reflect the customer's historical credit performance and stability degree; based on the short-term data, calculate the credit fluctuation coefficient to capture the customer's latest credit change characteristics; finally, calculate the static adjustment score and the dynamic adjustment score respectively based on the coefficients of these two dimensions to achieve a comprehensive evaluation of the customer's credit level; it not only ensures the stability of the evaluation results, but also improves the timeliness of risk warning.

[0019] In a second aspect, the present application provides a credit label determination system based on credit variables using AI, including: A credit feature vector acquisition module, which is used to acquire a credit feature vector representing the customer's credit status according to multi-dimensional data including customer information; A feature weight coefficient calculation module, which is used to calculate, based on the credit feature vector, a feature weight coefficient representing the influence degree of different credit features on credit evaluation by using a machine learning algorithm; A credit fluctuation coefficient calculation module, which is used to stratify and classify customers according to the feature weight coefficient and the credit score interval, and calculate the credit fluctuation coefficient of each type of customer, and the credit fluctuation coefficient represents the deviation degree between the customer's actual credit performance and the expected credit level; A credit label determination module, which is used to determine the final credit label of the customer through a comprehensive scoring model based on the feature weight coefficient, the credit fluctuation coefficient and the sample quantity of each type of customer, and the credit label reflects the overall credit level and risk degree of the customer.

[0020] Optionally, the credit feature vector acquisition module specifically includes: A transaction behavior feature extraction unit, configured to obtain historical transaction data of a customer, and extract transaction behavior features including transaction frequency, transaction scale, and transaction time distribution; A financial feature extraction unit, configured to obtain financial statement data of a customer, and extract financial features including asset scale, liability level, and cash flow status; A credit record feature extraction unit, configured to obtain credit investigation data of a customer, and extract credit record features including historical default records, overdue situations, and credit card usage status; An associated feature extraction unit, configured to obtain associated party data of a customer, and extract associated features including business conditions of associated enterprises, equity relationships, and industry credit levels; A feature fusion unit, configured to perform feature fusion on the transaction behavior features, financial features, credit record features, and associated features to obtain a credit feature vector representing the credit status of the customer.

[0021] Optionally, the feature weight coefficient calculation module specifically includes: A transaction behavior feature ranking unit, configured to rank the importance of the transaction behavior features, and calculate a transaction behavior time decay coefficient; A financial feature analysis unit, configured to perform industry grouping analysis on the financial features, and calculate an industry deviation coefficient; A credit record analysis unit, configured to perform default probability analysis on the credit record features, and calculate a default impact coefficient; An associated feature analysis unit, configured to perform associated conduction analysis on the associated features, and calculate an associated conduction coefficient; A weight coefficient calculation unit, configured to obtain feature weight coefficients of each credit feature based on the time decay coefficient, industry deviation coefficient, default impact coefficient, and associated conduction coefficient.

[0022] In summary, the present application includes at least one of the following beneficial technical effects: 1. The present application first collects and organizes multi-dimensional data including customer basic information, transaction behavior, social relationships, etc., and constructs a feature vector that can comprehensively represent the credit status of the customer; secondly, uses machine learning algorithms to calculate the influence weights of different credit features on credit assessment through learning of existing samples; then, stratifies customers based on feature weight coefficients and credit score intervals, and introduces a credit fluctuation coefficient to quantify the deviation degree of the actual credit performance of the customer from the expected level; finally, incorporates factors such as feature weights, credit fluctuation coefficients, and sample distributions of various types of customers into a comprehensive scoring model to scientifically determine the final credit label of the customer; through multi-dimensional data analysis and credit fluctuation coefficient design, more accurate credit assessment is achieved, and at the same time, the modeling cost and difficulty are significantly reduced; 2. This application first introduces the dimension of transaction behavior. By analyzing the transaction frequency, transaction scale, and time distribution characteristics of customers, it reveals the business activity and stability of customers. Secondly, it deeply analyzes the financial statement data, extracts key indicators such as asset scale, liability level, and cash flow status to evaluate the financial health of customers. Then, it integrates credit data, including historical default records, overdue situations, and credit card usage, to comprehensively reflect the credit performance ability of customers. Finally, it introduces the associated dimension analysis. By examining the business conditions of associated enterprises, the equity relationship network, and the credit level of the industry they belong to, it warns of potential transmission risks. Finally, it uses feature fusion technology to scientifically combine the features of the above four dimensions to construct a comprehensive and three-dimensional credit feature vector. Through the organic integration of multi-dimensional data, it significantly improves the comprehensiveness and accuracy of credit assessment and effectively avoids the risks brought by single-dimensional assessment. 3. This application first introduces the time dimension analysis. By ranking the importance of transaction behavior characteristics and calculating the time decay coefficient, it realizes the accurate characterization of recent behaviors. Secondly, it adopts the industry grouping analysis method. By calculating the deviation degree between financial characteristics and industry standards, it obtains the deviation coefficient reflecting industry characteristics. Then, it conducts default probability analysis. By statistically modeling historical credit records, it calculates the default impact coefficient to quantify the degree of credit risk. Finally, it constructs an associated transmission analysis framework. By evaluating the risk transmission effect between associated parties, it calculates the associated transmission coefficient to warn of potential risks. Finally, based on the coefficients of the above four dimensions, it is integrated to obtain a dynamically adjusted feature weight coefficient. The multi-dimensional dynamic weight calculation significantly improves the accuracy and timeliness of feature weights and effectively solves the limitations of traditional static weight methods. Brief Description of the Drawings

[0023] Figure 1 It is a schematic flowchart of a method for determining credit labels based on credit variables using AI according to an embodiment of this application; Figure 2 It is a schematic flowchart of step S100 in a method for determining credit labels based on credit variables using AI according to an embodiment of this application; Figure 3 It is a schematic flowchart of step S200 in a method for determining credit labels based on credit variables using AI according to an embodiment of this application; Figure 4 It is a schematic flowchart of step S300 in a method for determining credit labels based on credit variables using AI according to an embodiment of this application; Figure 5 It is a schematic flowchart of step S330 in a method for determining credit labels based on credit variables using AI according to an embodiment of this application; Figure 6It is a schematic flowchart of step S400 in a method for determining credit labels based on credit variables using AI according to an embodiment of the present application; Figure 7 It is a schematic flowchart of step S420 in a method for determining credit labels based on credit variables using AI according to an embodiment of the present application; Figure 8 It is a schematic diagram of modules of a system for determining credit labels based on credit variables using AI according to an embodiment of the present application. Detailed implementation manners

[0024] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification and appended claims of the present application, the singular forms "a", "an", "the", "above-mentioned", "said", and "this" are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present application refers to any or all possible combinations including one or more of the listed items.

[0025] Hereinafter, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as implying or suggesting relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present application, unless otherwise stated, the meaning of "a plurality" is two or more.

[0026] The following further describes the embodiments of the present application in conjunction with the drawings of the specification.

[0027] In a first aspect, the present application provides a method for determining credit labels based on credit variables using AI. Referring to Figure 1 , the method includes the following steps: S100. Obtain a credit feature vector characterizing the credit status of a customer according to multi-dimensional data including customer information.

[0028] In this embodiment, the multi-dimensional data is collected through various business terminals of the front-end application presentation layer, including customer basic information provided by the financing client and the core enterprise end, as well as transaction data recorded by business platforms such as factoring, micro-loans, and financial leasing; the credit feature vector refers to a set of features that can characterize the credit status of a customer.

[0029] Specifically, in this step, a customer portrait model is established relying on the business modeling engine of the technology middle platform, and data from each business system is obtained in real time through the integration engine. For structured data (such as enterprise scale, registered capital, etc.), it is directly extracted from the customer master data maintained by the unified user management engine; for unstructured data (such as credit reports), the document engine is used to parse and extract key information. For example, multi-dimensional data such as the historical repayment records of an enterprise in the microloan system and the transaction frequency in the factoring system can be obtained to form a complete feature vector.

[0030] S200. Based on the credit feature vector, use machine learning algorithms to calculate the feature weight coefficients that characterize the influence of different credit features on credit assessment.

[0031] In this embodiment, the feature weight coefficients reflect the importance of different credit features; the machine learning algorithms perform model training and weight calculation through the risk control engine of the business middle platform.

[0032] Specifically, a feature weight calculation rule library is established based on the rule engine of the technology middle platform. First, use the organizational structure engine to divide samples by business line, and train feature weight models for different business scenarios. In addition to considering objective data, the model can also combine the scoring opinions of each approver in the business approval process, and collect and quantify these subjective judgments through the process engine. Finally, the report engine regularly generates a feature weight analysis report and pushes it to the management personnel.

[0033] S300. According to the feature weight coefficients and credit score intervals, stratify and classify customers and calculate the credit fluctuation coefficient for each type of customer. The credit fluctuation coefficient characterizes the deviation degree of the actual credit performance of the customer from the expected credit level.

[0034] In this embodiment, the credit score interval is a preset scoring standard; the credit fluctuation coefficient characterizes the stability of credit performance.

[0035] Specifically, build a stratification and classification model in the business middle platform, and determine a reasonable stratification standard in combination with the historical data of each business system. Set different levels of approval authority through the system decentralization engine, and high-risk customers require a more stringent approval process. For customers with large score fluctuations, the message engine can timely push warning information to relevant personnel.

[0036] S400. Based on the feature weight coefficients, credit fluctuation coefficients, and the sample quantities of various types of customers, determine the final credit label of the customer through a comprehensive scoring model. The credit label reflects the overall credit level and risk degree of the customer.

[0037] In this embodiment, the comprehensive scoring model integrates multiple evaluation dimensions; the credit label is the final credit rating result.

[0038] Specifically, a unified credit rating display platform is established relying on the portal engine. The rating results are displayed according to different user roles through the permission engine. The management can view the overall distribution, while business personnel can only view the rating situations of the customers they are responsible for. At the same time, the operation and maintenance center monitors the operation of the model to ensure the timely update and accuracy of the rating results.

[0039] In one embodiment, the credit feature vector includes transaction behavior features, financial features, credit record features, and association features; referring to Figure 2 , in step S100, according to multi-dimensional data including customer information, a credit feature vector representing the customer's credit status is obtained, which specifically includes the following steps: S110. Obtain the historical transaction data of the customer, and extract transaction behavior features including transaction frequency, transaction scale, and transaction time distribution.

[0040] In this embodiment, the transaction behavior features refer to a set of indicators that can reflect the rules of the customer's transaction activities, including transaction frequency (the number of transactions within a unit time), transaction scale (the amount of a single transaction and the total amount), and transaction time distribution (the time pattern of transaction activity).

[0041] Specifically, in this step, a transaction feature extraction template library is first established, and the feature calculation rules under different business scenarios are preset. The system automatically extracts the transaction records of the past 1 year from the transaction flow database, statistically calculates the number of transactions per month to form a frequency index, and calculates the average monthly transaction amount to obtain a scale index. For the time distribution feature, a simple time period mapping table is established (such as dividing each day into three time periods: morning, afternoon, and evening), and the transaction proportion of the customer in each time period is statistically calculated.

[0042] S120. Obtain the financial statement data of the customer, and extract financial features including asset scale, debt level, and cash flow status.

[0043] In this embodiment, the financial features include various financial indicators reflecting the business conditions of the enterprise. The asset scale reflects the overall strength of the enterprise, the debt level reflects the debt repayment pressure, and the cash flow status indicates the short-term repayment ability.

[0044] Specifically, a financial index database is established, which contains the calculation formulas and industry standard values of various financial ratios. The system automatically extracts key data from the financial statements provided by the customer according to a unified accounting subject system. Basic indexes such as asset-liability ratio and current ratio are obtained through preset calculation rules. At the same time, an outlier identification rule table is established to mark the indexes that deviate significantly from the industry average.

[0045] S130. Obtain the credit investigation data of the customer, and extract credit record features including historical default records, overdue situations, and credit card usage status.

[0046] In this embodiment, the credit record feature reflects the customer's historical credit performance, including whether there is default, overdue days, credit card usage and repayment situation, etc.

[0047] Specifically, a credit information parsing rule library is established to extract standardized information from the credit report of the People's Bank. The system identifies negative information such as default and overdue through keyword matching, and sets simple scoring rules: for example, 100 points are counted for no default record, and points are deducted according to the number of overdue days. For the credit card usage situation, an evaluation matrix table is established, and scores are given by comprehensively considering the credit limit utilization rate and repayment timeliness.

[0048] S140. Obtain the related party data of the customer, and extract the related features including the operating conditions, equity relationship and industry credit level of the related enterprises.

[0049] In this embodiment, the related feature refers to the credit status of entities directly or indirectly related to the customer, including the operating conditions of related enterprises, the equity structure relationship, and the overall credit level of the industry to which they belong.

[0050] Specifically, first establish an enterprise-related graph database to record the equity, guarantee and other related relationships between enterprises. Obtain the equity structure information through the industrial and commercial data interface, and set the related degree calculation rule (for example, a shareholding ratio > 30% is recognized as a close relationship). Establish an industry credit rating table, and regularly update the credit risk levels of each industry as a reference benchmark for evaluating the credit status of individual enterprises.

[0051] S150. Perform feature fusion on the transaction behavior feature, financial feature, credit record feature and related feature to obtain a credit feature vector representing the customer's credit status.

[0052] In this embodiment, feature fusion refers to integrating the features of each dimension according to a unified standard to form a complete credit feature vector.

[0053] Specifically, establish a feature standardization conversion rule library to uniformly convert indicators with different dimensions into standard scores in the 0-1 interval. Use the linear weighted method for feature fusion, and the weights are set according to experience.

[0054] In one embodiment, with reference to Figure 3 , in step S200, based on the credit feature vector, use the machine learning algorithm to calculate the feature weight coefficients representing the influence of different credit features on credit evaluation, which specifically includes the following steps: S210. Sort the importance of the transaction behavior features and calculate the transaction behavior time decay coefficient.

[0055] In this embodiment, importance ranking refers to ranking according to the degree of influence of transaction behavior characteristics on credit assessment; the time decay coefficient is used to characterize the importance difference of transaction behaviors in different periods and reflects the timeliness characteristics of transaction data.

[0056] Specifically, establish a time weight mapping table for transaction characteristics and set a basic decay factor according to the time span. For example, divide the transaction data in the past year into 4 quarters. Calculate the transaction index scores for each quarter through the arithmetic average method, and multiply by the corresponding weights to obtain the weighted average. For industries with obvious seasonality, set a seasonal adjustment coefficient in the mapping table. For example, the transaction fluctuations during holidays in the retail industry will be smoothed. The system regularly runs a time series analysis program to identify the trend characteristics of transaction data.

[0057] S220. Conduct industry grouping analysis on financial characteristics and calculate the industry deviation coefficient.

[0058] In this embodiment, industry grouping analysis refers to dividing enterprises in the same or similar industries into the same group for comparative analysis; the industry deviation coefficient reflects the degree of difference between an enterprise's financial indicators and the industry average level.

[0059] S230. Conduct default probability analysis on credit record characteristics and calculate the default impact coefficient.

[0060] In this embodiment, default probability analysis refers to evaluating the possibility of future default based on historical credit records; the default impact coefficient represents the impact weight of different types of default behaviors on credit assessment.

[0061] Specifically, establish a default behavior classification library, classify default events according to dimensions such as nature and amount, and set corresponding basic impact scores.

[0062] S240. Conduct association conduction analysis on association characteristics and calculate the association conduction coefficient.

[0063] In this embodiment, association conduction analysis refers to evaluating the degree of influence of the credit status of associated parties on the target enterprise; the association conduction coefficient reflects the conduction intensity of this credit influence.

[0064] Specifically, establish an association relationship evaluation matrix, and set a basic conduction coefficient according to the type of association relationship (such as equity, guarantee, etc.) and the degree of association. For example, the conduction coefficient for a direct holding relationship is 0.8, and the conduction coefficient for a general business transaction relationship is 0.2. If a major credit risk event occurs to an associated party, then adjust the credit assessment result of the target enterprise accordingly according to the conduction coefficient. For enterprises within a group, considering the overall credit linkage, appropriately increase the conduction coefficient.

[0065] S250. Based on the time decay coefficient, industry deviation coefficient, default impact coefficient and correlation transmission coefficient, the characteristic weight coefficient of each credit characteristic is obtained.

[0066] In this embodiment, the feature weight coefficient refers to the importance weight of each type of credit feature in the final credit assessment, which is used to integrate the assessment results of various dimensions.

[0067] Specifically, a feature weight calculation template is established and the benchmark weight value of each coefficient is set.

[0068] In one embodiment, referring to Figure 4 In step S300, according to the characteristic weight coefficient and the credit score range, the customers are stratified and classified and the credit volatility coefficient of each type of customer is calculated, which specifically includes the following steps: S310. Classify customer samples based on feature weight coefficients and credit score ranges, and divide customers into different credit rating categories.

[0069] In this embodiment, the credit rating category refers to the hierarchical division based on the quality of the customer's credit status; the classification standard needs to comprehensively consider the feature weights and score ranges to ensure the rationality and explainability of the category division.

[0070] S320. Calculate the historical credit score sequence of customers in each credit rating category to obtain the credit score fluctuation range within the category.

[0071] In this embodiment, the historical credit score sequence refers to the record of changes in the credit score of a customer within a certain period of time; the credit score fluctuation range represents the score difference range of customers in the same category.

[0072] Specifically, we build a scoring tracking database to record monthly scoring data to form a time series. We group and count by credit rating category, and calculate the concentration trend and dispersion of scores in each category. We build a statistical analysis template to calculate the mean and standard deviation of monthly scores to obtain the normal fluctuation range. For outliers that significantly deviate from the normal range, we build an early warning marking mechanism to promptly identify customers with abnormal score fluctuations.

[0073] S330. Analyze the fluctuation trajectory of each customer within the scoring range, calculate the deviation between the fluctuation trajectory and the category fluctuation range, and obtain the credit stability index of the individual customer.

[0074] In this embodiment, the fluctuation trajectory refers to the path characteristics of customer ratings changing over time; the deviation degree indicates the degree of difference between the fluctuation of individual ratings and the overall fluctuation pattern of the category to which it belongs.

[0075] S340. Count the distribution of stability indicators of customers in each credit rating category and determine the overall volatility characteristics of the category.

[0076] In this embodiment, the stability index distribution reflects the score fluctuation of the customer group within the same credit rating category; the category fluctuation characteristic is a comprehensive measure of the stability of the entire category.

[0077] S350. Calculate the credit volatility coefficient of each type of customer based on the stability index of the individual customer and the category volatility characteristics.

[0078] In this embodiment, the credit volatility coefficient is an indicator that comprehensively reflects the stability of the customer's credit score, and it is necessary to consider the volatility characteristics at both the individual and category levels.

[0079] Specifically, a volatility coefficient calculation rule base is established, and a basic calculation formula is set. The weighted average method is used to comprehensively calculate the individual stability index and the category volatility characteristics to obtain the final volatility coefficient.

[0080] In one embodiment, referring to Figure 5 In step S330, the fluctuation trajectory of each customer within the scoring interval is analyzed, and the deviation between the fluctuation trajectory and the category fluctuation range is calculated, which specifically includes the following steps: S331. Obtain the customer's continuous rating data during the evaluation period, and calculate the maximum, minimum and mean values of the rating sequence.

[0081] In this embodiment, continuous scoring data refers to the monthly credit score records of customers within a fixed evaluation period (such as 12 months); the statistical characteristics of the scoring sequence (maximum value, minimum value, mean) are used to characterize the fluctuation range and central tendency of the score.

[0082] Specifically, a scoring sequence analysis template is established. First, the original scoring data is cleaned and obvious outliers are removed.

[0083] S332. Set fluctuation tolerance parameters according to the credit level of the credit rating category.

[0084] Among them, the higher the credit level, the smaller the volatility tolerance parameter, and the lower the credit level, the larger the volatility tolerance parameter.

[0085] In this embodiment, the volatility tolerance parameter refers to the acceptable volatility range standard set for customers with different credit ratings; this parameter has an inverse relationship with the credit level, reflecting more stringent stability requirements for high-credit customers. From the perspective of risk sensitivity, customers with a high credit level usually have stable operations and good credit records. Therefore, even a small credit fluctuation may indicate a major risk signal. Setting a small volatility tolerance parameter helps to detect potential risks at an early stage; from a business logic perspective, customers with a high credit level enjoy more favorable credit terms, and correspondingly, higher stability requirements should be imposed on them; from the perspective of risk management requirements, the default of customers with a high credit level may cause greater losses, and stricter monitoring standards are needed to protect credit assets. A small volatility tolerance parameter helps to improve the sensitivity of risk warning; from practical experience, customers with a low credit level already have large fluctuations. Therefore, it is more in line with the actual situation to adopt a larger volatility tolerance parameter, avoiding misjudgment caused by excessive sensitivity.

[0086] S333. Calculate the upper and lower limits of the volatility benchmark for each credit rating category based on the category volatility tolerance parameter and historical volatility data.

[0087] In this embodiment, the upper and lower limits of the volatility benchmark define the range boundaries of normal fluctuations for each credit rating category; the benchmark interval needs to comprehensively consider the historical data performance and volatility tolerance requirements.

[0088] S334. Calculate the overlap degree and the exceeding degree between the volatility interval of the customer score sequence and the corresponding category volatility benchmark interval respectively to obtain the deviation degree considering the volatility tolerance.

[0089] In this embodiment, the overlap degree represents the intersection ratio between the customer score volatility interval and the benchmark interval; the exceeding degree represents the degree to which the volatility exceeds the benchmark interval; these two indicators jointly constitute the measurement standard of the deviation degree.

[0090] Specifically, establish a deviation degree calculation rule library and set the standardization method for score interval comparison. Calculate the overlap degree by dividing the length of the overlapping interval by the length of the benchmark interval; calculate the exceeding degree by dividing the number of points exceeding the benchmark interval by the total number of observation points. Add the overlap degree score and the deduction value of the exceeding degree to obtain the final deviation degree score.

[0091] In one embodiment, referring to Figure 6 , in step S400, based on the feature weight coefficient, the credit volatility coefficient, and the sample quantity of various types of customers, determine the final credit label of the customer through a comprehensive scoring model, which specifically includes the following steps: S410. Multiply the credit characteristics of the customer by the corresponding feature weight coefficient to obtain the weighted feature score.

[0092] S420. Modify the weighted feature scores according to the credit fluctuation coefficient of the customer's category to obtain an adjusted score considering volatility.

[0093] In this embodiment, the adjusted score refers to the modified score after considering the impact of credit fluctuations; adjusting the original score through the fluctuation coefficient can better reflect the stability characteristics of the customer's credit.

[0094] S430. Calculate the sample sparsity coefficient based on the sample quantity distribution of various types of customers.

[0095] S440. Statistically analyze the score distribution characteristics of each credit rating category to determine the classification threshold of the credit label.

[0096] In this embodiment, the classification threshold is the critical point for distinguishing different credit labels; the score distribution characteristics include the central tendency and dispersion degree of the scores of each level.

[0097] S450. Determine the final credit label of the customer according to the adjusted score, sample sparsity coefficient, and classification threshold of the customer.

[0098] In one embodiment, referring to Figure 7 , the adjusted score includes a static adjusted score and a dynamic adjusted score. In step S420, the acquisition of the adjusted score specifically includes the following steps: S421. Divide the customer's historical data into a long-term evaluation window and a short-term evaluation window according to the time dimension, which are respectively used to calculate the static credit level and dynamic credit trend.

[0099] S422. Calculate the long-term credit stability coefficient and short-term credit fluctuation coefficient respectively according to the historical evaluation data.

[0100] In this embodiment, the long-term evaluation window is used to capture the customer's basic credit level and stability characteristics, while the short-term evaluation window focuses on identifying the recent credit change trend and fluctuation characteristics. This dual-window mechanism can not only reflect the customer's credit fundamentals but also timely detect short-term changes in credit risks.

[0101] S423. Obtain the static adjusted score and dynamic adjusted score of the customer respectively according to the long-term credit stability coefficient and short-term credit fluctuation coefficient.

[0102] Specifically, through the comparative analysis of the long-term credit stability coefficient and short-term credit fluctuation coefficient, abnormal changes in the customer's credit status can be effectively identified, especially early warning signals of credit decline can be detected in advance.

[0103] It should be understood that the sequence numbers of the steps in the above embodiments do not indicate the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0104] In a second aspect, the present application provides a credit label determination system using AI based on credit variables. Below, in combination with the above-mentioned credit label determination method using AI based on credit variables, the credit label determination system using AI based on credit variables of the present application will be described.

[0105] Refer to Figure 8 , a credit label determination system using AI based on credit variables, includes: A credit feature vector acquisition module, configured to acquire a credit feature vector representing the credit status of a customer according to multi-dimensional data including customer information; A feature weight coefficient calculation module, configured to calculate, based on the credit feature vector, a feature weight coefficient representing the influence degree of different credit features on credit evaluation by using a machine learning algorithm; A credit fluctuation coefficient calculation module, configured to stratify and classify customers according to the feature weight coefficient and the credit score interval, and calculate the credit fluctuation coefficient of each type of customer. The credit fluctuation coefficient represents the deviation degree of the actual credit performance of the customer from the expected credit level; A credit label determination module, configured to determine the final credit label of the customer through a comprehensive scoring model based on the feature weight coefficient, the credit fluctuation coefficient, and the sample quantity of each type of customer. The credit label reflects the overall credit level and risk degree of the customer.

[0106] In one embodiment, the credit feature vector acquisition module specifically includes: A transaction behavior feature extraction unit, configured to acquire the historical transaction data of the customer, and extract transaction behavior features including transaction frequency, transaction scale, and transaction time distribution; A financial feature extraction unit, configured to acquire the financial statement data of the customer, and extract financial features including asset scale, liability level, and cash flow status; A credit record feature extraction unit, configured to acquire the credit investigation data of the customer, and extract credit record features including historical default records, overdue situations, and credit card usage status; An associated feature extraction unit, configured to acquire the associated party data of the customer, and extract associated features including the business status of associated enterprises, equity relationships, and industry credit levels; A feature fusion unit, configured to perform feature fusion on the transaction behavior features, financial features, credit record features, and associated features to obtain a credit feature vector representing the credit status of the customer.

[0107] In one embodiment, the feature weight coefficient calculation module specifically includes: A transaction behavior feature sorting unit, configured to sort the importance of transaction behavior features and calculate a transaction behavior time decay coefficient; A financial feature analysis unit, configured to perform industry grouping analysis on financial features and calculate an industry deviation coefficient; A credit record analysis unit, configured to perform default probability analysis on credit record features and calculate a default impact coefficient; An associated feature analysis unit, configured to perform associated conduction analysis on associated features and calculate an associated conduction coefficient; A weight coefficient calculation unit, configured to obtain the feature weight coefficients of each credit feature based on the time decay coefficient, industry deviation coefficient, default impact coefficient, and associated conduction coefficient.

[0108] In one embodiment, the credit fluctuation coefficient calculation module specifically includes: A customer classification unit, configured to classify customer samples by combining the feature weight coefficients and credit score intervals, and divide customers into different credit rating categories; A historical score analysis unit, configured to calculate the historical credit score sequences of customers within each credit rating category and obtain the credit score fluctuation range within the category; A fluctuation trajectory analysis unit, configured to analyze the fluctuation trajectories of each customer within the score interval, calculate the deviation degree between the fluctuation trajectory and the category fluctuation range, and obtain the credit stability index of the customer individual; A fluctuation feature statistics unit, configured to statistically analyze the distribution of stability indexes of customers within each credit rating category and determine the overall fluctuation characteristics of the category; A fluctuation coefficient calculation unit, configured to calculate the credit fluctuation coefficients of each type of customer based on the stability index of the customer individual and the category fluctuation characteristics.

[0109] In one embodiment, the fluctuation trajectory analysis unit specifically includes: A score data extraction unit, configured to obtain the continuous score data of a customer during the evaluation period and calculate the maximum value, minimum value, and average value of the score sequence; A fluctuation tolerance parameter setting unit, configured to set a fluctuation tolerance parameter according to the credit level of the credit rating category. The higher the credit level, the smaller the fluctuation tolerance parameter, and the lower the credit level, the larger the fluctuation tolerance parameter; A fluctuation benchmark calculation unit, configured to calculate the upper and lower limits of the fluctuation benchmark for each credit rating category based on the category fluctuation tolerance parameter and historical fluctuation data; A deviation degree calculation unit, configured to calculate the overlap degree and exceed degree between the fluctuation interval of the customer score sequence and the corresponding category fluctuation benchmark interval, and obtain the deviation degree considering the fluctuation tolerance.

[0110] In one embodiment, the credit label determination module specifically includes: A weighted score calculation unit, configured to multiply the credit characteristics of a customer by the corresponding characteristic weight coefficient to obtain a weighted characteristic score; An adjusted score correction unit, configured to correct the weighted characteristic score according to the credit fluctuation coefficient of the category to which the customer belongs to obtain an adjusted score considering volatility; A sample sparsity calculation unit, configured to calculate a sample sparsity coefficient based on the sample quantity distribution of various types of customers; A classification threshold determination unit, configured to count the score distribution characteristics of each credit rating category and determine the classification threshold of the credit label; A credit label assignment unit, configured to determine the final credit label of a customer according to the adjusted score, sample sparsity coefficient, and classification threshold of the customer.

[0111] In one embodiment, the adjusted score correction unit specifically includes: An evaluation window division unit, configured to divide the historical data of a customer into a long-term evaluation window and a short-term evaluation window according to the time dimension, respectively for calculating the static credit level and dynamic credit trend; A stability coefficient calculation unit, configured to calculate a long-term credit stability coefficient according to historical evaluation data; A fluctuation coefficient calculation unit, configured to calculate a short-term credit fluctuation coefficient according to historical evaluation data; An adjusted score calculation unit, configured to obtain the static adjusted score and dynamic adjusted score of a customer according to the long-term credit stability coefficient and short-term credit fluctuation coefficient, respectively.

[0112] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for determining credit labels based on credit variables using AI, characterized in that, It includes the following steps: Based on multi-dimensional data including customer information, obtain a credit feature vector characterizing the customer's credit status; Based on the credit feature vector, use a machine learning algorithm to calculate feature weight coefficients characterizing the influence of different credit features on credit assessment; According to the feature weight coefficients and credit score intervals, stratify and classify customers and calculate the credit fluctuation coefficient for each category of customers. The credit fluctuation coefficient characterizes the deviation degree of the customer's actual credit performance from the expected credit level; Based on the feature weight coefficients, credit fluctuation coefficients, and the sample sizes of various categories of customers, determine the final credit label of the customer through a comprehensive scoring model. The credit label reflects the overall credit level and risk degree of the customer.

2. The method for determining credit labels based on credit variables using AI according to claim 1, wherein The credit feature vector includes transaction behavior features, financial features, credit record features, and association features. Based on multi-dimensional data including customer information, obtaining a credit feature vector characterizing the customer's credit status specifically includes the following steps: Obtain the customer's historical transaction data and extract transaction behavior features including transaction frequency, transaction scale, and transaction time distribution; Obtain the customer's financial statement data and extract financial features including asset scale, liability level, and cash flow status; Obtain the customer's credit investigation data and extract credit record features including historical default records, overdue situations, and credit card usage status; Obtain the customer's related party data and extract association features including the business conditions of related enterprises, equity relationships, and industry credit levels; Perform feature fusion on the transaction behavior features, financial features, credit record features, and association features to obtain a credit feature vector characterizing the customer's credit status.

3. The method for determining credit labels based on credit variables using AI according to claim 2, wherein Based on the credit feature vector, use a machine learning algorithm to calculate feature weight coefficients characterizing the influence of different credit features on credit assessment. Specifically, it includes the following steps: Rank the importance of the transaction behavior features and calculate the transaction behavior time decay coefficient; Conduct industry grouping analysis on the financial features and calculate the industry deviation coefficient; Conduct default probability analysis on the credit record features and calculate the default impact coefficient; Conduct association conduction analysis on the association features and calculate the association conduction coefficient; Based on the time decay coefficient, industry deviation coefficient, default impact coefficient, and association conduction coefficient, obtain the feature weight coefficients of each credit feature.

4. The method for determining credit labels based on credit variables using AI according to claim 1, wherein According to the feature weight coefficients and credit score intervals, stratify and classify customers and calculate the credit fluctuation coefficient for each category of customers. Specifically, it includes the following steps: Combine the feature weight coefficients and credit score intervals to classify the customer samples and divide the customers into different credit grade categories; Calculate the historical credit score sequence of customers within each credit grade category to obtain the credit score fluctuation range within the category; Analyze the fluctuation trajectory of each customer within the score interval, calculate the deviation degree of the fluctuation trajectory from the category fluctuation range, and obtain the credit stability index of the customer individual; Statistically analyze the distribution of the stability indices of customers within each credit grade category to determine the overall fluctuation characteristics of the category; Based on the stability index of the customer individual and the category fluctuation characteristics, calculate the credit fluctuation coefficient for each category of customers.

5. The method for determining credit labels based on credit variables using AI according to claim 4, characterized in that, Analyze the fluctuation trajectories of each customer within the rating range, and calculate the deviation degree of the fluctuation trajectory from the category fluctuation range. The specific steps are as follows: Obtain the continuous rating data of the customer during the evaluation period, and calculate the maximum value, minimum value, and average value of the rating sequence; Set the fluctuation tolerance parameter according to the credit level of the credit rating category. The higher the credit level, the smaller the fluctuation tolerance parameter; the lower the credit level, the larger the fluctuation tolerance parameter; Based on the category fluctuation tolerance parameter and historical fluctuation data, calculate the upper and lower limits of the fluctuation benchmark for each credit rating category; Calculate the overlap degree and exceedance degree between the fluctuation range of the customer rating sequence and the corresponding category fluctuation benchmark range respectively, and obtain the deviation degree considering the fluctuation tolerance; 6. The method for determining credit labels based on credit variables using AI according to claim 1, wherein Based on the feature weight coefficient, credit fluctuation coefficient, and the sample quantity of each type of customer, determine the final credit label of the customer through a comprehensive scoring model. The specific steps are as follows: Multiply the credit characteristics of the customer by the corresponding feature weight coefficient to obtain the weighted feature score; According to the credit fluctuation coefficient of the category to which the customer belongs, correct the weighted feature score to obtain the adjusted score considering volatility; Based on the sample quantity distribution of each type of customer, calculate the sample sparsity coefficient; Statistically analyze the score distribution characteristics of each credit rating category, and determine the classification threshold of the credit label; Determine the final credit label of the customer according to the adjusted score, sample sparsity coefficient, and classification threshold of the customer; 7. The method for determining credit labels based on credit variables using AI according to claim 6, characterized in that, The adjusted score includes a static adjusted score and a dynamic adjusted score. The acquisition of the adjusted score specifically includes the following steps: Divide the historical data of the customer into a long-term evaluation window and a short-term evaluation window according to the time dimension, which are respectively used to calculate the static credit level and dynamic credit trend; According to the historical evaluation data, calculate the long-term credit stability coefficient and the short-term credit fluctuation coefficient respectively; According to the long-term credit stability coefficient and the short-term credit fluctuation coefficient, obtain the static adjusted score and the dynamic adjusted score of the customer respectively; 8. A credit label determination system using AI based on credit variables, characterized in that, Include: A credit feature vector acquisition module, which is used to obtain a credit feature vector representing the credit status of the customer according to multi-dimensional data including customer information; A feature weight coefficient calculation module, which is used to calculate, based on the credit feature vector, a feature weight coefficient representing the influence degree of different credit features on credit evaluation by using a machine learning algorithm; A credit fluctuation coefficient calculation module, which is used to classify and stratify the customers according to the feature weight coefficient and credit score range, and calculate the credit fluctuation coefficient of each type of customer. The credit fluctuation coefficient represents the deviation degree between the actual credit performance of the customer and the expected credit level; A credit label determination module, which is used to determine the final credit label of the customer through a comprehensive scoring model based on the feature weight coefficient, credit fluctuation coefficient, and the sample quantity of each type of customer. The credit label reflects the overall credit level and risk degree of the customer; 9. The credit label determination system using AI based on credit variables according to claim 8, characterized in that, The credit feature vector acquisition module specifically includes: A transaction behavior feature extraction unit, which is used to obtain the historical transaction data of the customer and extract transaction behavior features including transaction frequency, transaction scale, and transaction time distribution; A financial feature extraction unit for obtaining the financial statement data of a customer and extracting financial features including asset size, liability level, and cash flow status; A credit record feature extraction unit for obtaining the credit investigation data of a customer and extracting credit record features including historical default records, overdue situations, and credit card usage status; An associated feature extraction unit for obtaining the associated party data of a customer and extracting associated features including the business conditions of associated enterprises, equity relationships, and industry credit levels; A feature fusion unit for fusing the transaction behavior features, financial features, credit record features, and associated features to obtain a credit feature vector characterizing the credit status of the customer.

10. The credit label determination system using AI based on credit variables according to claim 8, characterized in that, The feature weight coefficient calculation module specifically includes: A transaction behavior feature sorting unit for sorting the importance of the transaction behavior features and calculating a transaction behavior time decay coefficient; A financial feature analysis unit for performing industry grouping analysis on the financial features and calculating an industry deviation coefficient; A credit record analysis unit for performing default probability analysis on the credit record features and calculating a default impact coefficient; An associated feature analysis unit for performing associated conduction analysis on the associated features and calculating an associated conduction coefficient; A weight coefficient calculation unit for obtaining the feature weight coefficients of each credit feature based on the time decay coefficient, industry deviation coefficient, default impact coefficient, and associated conduction coefficient.

Images