Vehicle loan risk prediction method and apparatus, storage medium, and electronic device

Abstract

The present application relates to the field of machine learning, and provides a vehicle loan risk prediction method and apparatus, a storage medium, and an electronic device. The method comprises: an electronic device acquiring an expiration risk feature of a borrower in a current repayment period (S1); processing the expiration risk feature by means of a risk assessment model of a neural network architecture to obtain a first risk score of the borrower in the current repayment period (S2); processing the expiration risk feature by means of a pre-fitted autoregressive integrated moving average model to obtain a second risk score of the borrower in the current repayment period (S3); and obtaining a comprehensive risk score of the borrower on the basis of the first risk score and the second risk score (S4). A risk assessment model of a neural network architecture can better capture the impacts of a sudden event, so that the defect that an autoregressive integrated moving average model is insensitive to a sudden event is overcome, thereby improving the expiration risk prediction accuracy.

Description

Vehicle loan risk prediction method and device, storage medium and electronic equipment

[0001] Cross-reference to Related Applications

[0002] The present application claims priority to the Chinese patent application No. 2025102181777, filed on February 26, 2025, and entitled "Vehicle loan risk prediction method and device, storage medium and electronic equipment", the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0003] The present application relates to the field of machine learning, in particular, to a vehicle loan risk prediction method and device, storage medium and electronic equipment. BACKGROUND

[0004] In modern financial risk management, predicting risk is a crucial task. Traditional risk prediction models are mainly based on historical data and fixed risk features for modeling, and then use these models to predict future risks. However, this method has some obvious shortcomings. Since this method usually assumes that risk features are static, it ignores the fact that risk features can change over time. Even though some techniques have tried to improve the model by adapting to changes in risk features, these improved methods are still not sensitive enough to sudden risks.

[0005] SUMMARY

[0006] In order to overcome at least one of the deficiencies in the prior art, the present application provides a vehicle loan risk prediction method and device, storage medium and electronic equipment, specifically including:

[0007] In a first aspect, the present application provides a vehicle loan risk prediction method, the method comprising:

[0008] obtaining overdue risk features of a borrower in a current repayment period;

[0009] processing the overdue risk features through a risk assessment model of a neural network architecture to obtain a first risk score of the borrower in the current repayment period;

[0010] processing the overdue risk features through a pre-fitted autoregressive moving average model to obtain a second risk score of the borrower in the current repayment period;

[0011] obtaining a comprehensive risk score of the borrower according to the first risk score and the second risk score.

[0012] In a second aspect, the present application provides a vehicle loan risk prediction device, the device comprising:

[0013] The risk characteristics module is used to obtain the borrower's delinquency risk characteristics in the current repayment period;

[0014] The risk scoring module is used to process the overdue risk characteristics through a risk assessment model with a neural network architecture to obtain the borrower's first risk score in the current repayment period;

[0015] The risk scoring module is also used to process the overdue risk characteristics through a pre-fitted autoregressive moving average model to obtain the borrower's second risk score in the current repayment period.

[0016] The risk integration module is used to obtain the borrower's comprehensive risk score based on the first risk score and the second risk score.

[0017] Thirdly, this application provides a storage medium storing a computer program, which, when processed by a processor, implements the aforementioned car loan risk prediction method.

[0018] Fourthly, this application provides an electronic device, which includes a processor and a memory. The memory stores a computer program, and the computer program, when processed by the processor, implements the aforementioned car loan risk prediction method.

[0019] Compared with the prior art, this application has the following beneficial effects:

[0020] This application provides a method, apparatus, storage medium, and electronic device for predicting auto loan risk. The electronic device acquires the borrower's delinquency risk characteristics for the current repayment period; processes these characteristics using a risk assessment model with a neural network architecture to obtain a first risk score for the borrower in the current repayment period; processes these characteristics again using a pre-fitted autoregressive moving average model to obtain a second risk score; and finally, based on the first and second risk scores, obtains the borrower's comprehensive risk score. Thus, because the risk assessment model with a neural network architecture can better capture the impact of sudden events, it compensates for the autoregressive moving average model's insensitivity to sudden events, thereby improving the accuracy of delinquency risk prediction. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 is a flowchart illustrating the auto loan risk prediction method provided in an embodiment of this application;

[0023] Figure 2 is one of the detailed flowcharts of the car loan risk prediction method provided in the embodiments of this application;

[0024] Figure 3 is a schematic diagram of feature change information provided in an embodiment of this application;

[0025] Figure 4 is a second detailed flowchart of the car loan risk prediction method provided in the embodiments of this application;

[0026] Figure 5 is a detailed flowchart of the car loan risk prediction method provided in the embodiments of this application (Part 3).

[0027] Figure 6 is a schematic diagram of the structure of the auto loan risk prediction device provided in the embodiment of this application;

[0028] Figure 7 is a schematic diagram of the structure of the electronic device provided in the embodiment of this application. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0030] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0031] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0032] In the description of this application, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0033] Based on the above statement, as described in the background section, even though some techniques have attempted to improve models by adapting to changes in risk characteristics, these improvement methods are still not sensitive enough to sudden risks.

[0034] For example, let's take the Autoregressive Integrated Moving Average (ARIMA) model (hereinafter referred to as the ARIMA model). The ARIMA model is a commonly used time series forecasting method that combines an autoregressive (AR) model and a moving average (MA) model. The basic idea of ​​the ARIMA model is to analyze historical time series data, identify patterns and trends, and use these patterns to predict future values. To achieve this, the ARIMA model comprises three components: an autoregressive term (AR), a differencing term (I), and a moving average term (MA). The autoregressive term represents the linear relationship between the current value and values ​​from several past periods; the differencing term is used to smooth the time series; and the moving average term is used to capture the autocorrelation of the error term. Through the combination of these three parts, the ARIMA model can effectively capture short-term fluctuations and long-term trends in time series.

[0035] However, while the ARIMA model performs well in handling stationary time series data, it is not sensitive enough to sudden risks. This is mainly because the ARIMA model relies heavily on patterns and trends in historical data for prediction. When sudden events occur (such as economic crises, natural disasters, etc.), these events are often unprecedented and cannot be predicted using historical data. Therefore, the ARIMA model may not be able to adjust its predictive model in time to cope with these sudden changes, leading to inaccurate predictions.

[0036] Based on the discovery of the aforementioned technical problems, the inventors, through creative labor, proposed the following technical solutions to solve or improve these problems. It should be noted that the deficiencies in the solutions of the prior art are the result of the inventors' practical experience and careful research. Therefore, the discovery process of the aforementioned problems and the solutions proposed in the embodiments of this application below should be considered contributions made by the inventors to this application during the inventive process, and should not be construed as technical content known to those skilled in the art.

[0037] In view of the above problems, this embodiment provides a method for predicting auto loan risks. As shown in Figure 1, the method includes:

[0038] S1 obtains the borrower's delinquency risk characteristics for the current repayment period.

[0039] S2 uses a risk assessment model with a neural network architecture to process the characteristics of overdue risk and obtain the borrower's first risk score for the current repayment period.

[0040] S3 uses a pre-fitted autoregressive moving average model to process the delinquency risk characteristics and obtain the borrower's second risk score for the current repayment period.

[0041] S4. Based on the first risk score and the second risk score, the borrower's overall risk score is obtained.

[0042] Thus, because the risk assessment model based on the neural network architecture can better capture the impact of sudden events, it makes up for the insensitivity of the autoregressive moving average model to sudden events and improves the prediction accuracy of overdue risk.

[0043] It is worth noting that, for the solutions involved in this application, all data related to user privacy was explicitly disclosed and authorized by the user before being collected or generated during the execution of the embodiments of this application. Furthermore, the location of the storage device for the personal information involved in the embodiments of this application complies with the legal requirements of the country / region where the aforementioned data-related activities occurred, including but not limited to: authorization, generation, use, and storage. It can be understood that the location of the storage device for the overdue risk feature used in implementing this solution complies with the legal requirements of the country / region where the aforementioned data-related activities occurred, including but not limited to: authorization, generation, use, and storage.

[0044] Furthermore, it should be understood that the electronic device implementing the car loan risk prediction method of this embodiment can be, but is not limited to, a mobile terminal, tablet computer, laptop computer, desktop computer, and server. The server can be a single server or a group of servers. The server group can be centralized or distributed (e.g., the server can be a distributed system). In some embodiments, the server can be local or remote relative to the user terminal. In some embodiments, the server can be implemented on a cloud platform; by way of example only, the cloud platform can include private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, inter-cloud, multi-cloud, etc., or any combination thereof. In some embodiments, the server can be implemented on an electronic device having one or more components.

[0045] To make the solution provided in this embodiment clearer, the following describes in detail each step of the method shown in Figure 1, using a server as the electronic device implementing the method. However, it should be understood that the operations in the flowchart may not be implemented in sequence, and steps without logical contextual relationships may be reversed in order or implemented simultaneously. Furthermore, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowchart, or remove one or more operations from the flowchart. Therefore, continuing to refer to Figure 1, the method includes:

[0046] S1 obtains the borrower's delinquency risk characteristics for the current repayment period.

[0047] In this embodiment, the aforementioned multiple delinquency sub-features include dynamic and static features that reflect whether a borrower will default. For example, the multiple delinquency sub-features may include the borrower's credit history, income stability, vehicle-related features, economic environment risk features, user personal features, and vehicle location information. These features are described in detail below:

[0048] (1) Borrower’s credit history characteristics, including credit card overdraft records, repayment records of other loans, credit score, overdue repayment records, and default records.

[0049] Credit card overdraft records reflect a borrower's credit card usage, including whether there were overdrafts, the frequency of overdrafts, and the amount. If a borrower frequently overdrafts and fails to repay on time, it indicates poor financial management skills and a potential for default on car loan repayments.

[0050] Other loan repayment records include the borrower's repayment records for all previous loans (such as housing loans, consumer loans, etc.). If the borrower has repeatedly defaulted on payments or defaulted on loans, it indicates a low willingness to repay and a higher likelihood of defaulting on future car loan repayments.

[0051] A credit score is a quantitative indicator used to assess a borrower's creditworthiness. It is typically calculated by credit scoring agencies based on a borrower's historical repayment records, overdraft history, and other factors. A higher credit score indicates better creditworthiness and a lower risk of default; conversely, a lower credit score implies a higher risk of default.

[0052] Default history is a feature that focuses on whether a borrower has defaulted on previous loans. For example, if a customer has accumulated [X] overdue credit card payments in the two years prior to applying for a car loan, with the longest consecutive overdue payment lasting [X] months, financial institutions can preliminarily determine that this customer has an extremely high risk of defaulting on the car loan.

[0053] Therefore, a comprehensive assessment of a borrower's historical repayment records, overdraft status, credit score, etc., can effectively identify potential default risks.

[0054] (2) Income stability characteristics are another important component of the risk assessment of car loans in this embodiment. They are mainly used to examine factors such as the borrower's occupation type, years of service, and source of income in order to determine the stability and reliability of their income.

[0055] Occupational types vary significantly due to differences in income stability and reliability. For example, civil servants, employees of state-owned enterprises, and formal employees of large corporations have relatively lower default risks compared to freelancers, self-employed individuals, or small business owners because of the stability and reliability of their jobs and income. These occupations typically offer fixed salaries and benefits, reducing the risk of default due to income instability.

[0056] Years of employment: Borrowers with longer employment histories typically have greater income stability and job security. In contrast, borrowers with shorter employment histories may face greater income volatility and uncertainty. For example, borrowers with more than [X] years of employment have significantly greater income stability than borrowers with less than [X] years of employment.

[0057] Besides occupation and length of employment, income source is also a crucial factor in measuring income stability. For example, borrowers with multiple stable income sources (such as fixed salary, rental income, investment returns, etc.) have higher income stability compared to borrowers with only a single income source. Diversified income sources can effectively mitigate the risks associated with a single income source.

[0058] Analyzing income trends can provide further insight into the stability of a borrower's income over the past few years. If a borrower's income has been on an upward trend or remained stable over the past few years, this indicates strong income stability and repayment ability; conversely, if income has been declining, the risk of default is higher.

[0059] Career prospects: Certain industries offer better career prospects and greater potential for income growth. For example, the high-tech and financial industries typically offer good career development opportunities and high income levels. Therefore, borrowers in these industries usually have higher income stability and repayment ability.

[0060] Therefore, this embodiment can effectively identify a borrower's income stability and repayment ability by comprehensively assessing the borrower's occupation type, years of service, income source, income trend, and career prospects.

[0061] (3) Vehicle-related characteristics are an important part of car loan risk assessment, mainly involving information such as the vehicle's value, purpose, and purchase channel, which can assess the value and reliability of the vehicle as collateral.

[0062] Vehicle value, specifically its market value, is a key factor in determining car loan amounts and loan-to-value ratios. Research shows that the higher the market value of a vehicle, the safer it is as collateral. Financial institutions typically refer to vehicle valuation reports from third-party appraisal agencies to determine the vehicle's actual value. Furthermore, the vehicle's residual value is also an important consideration; vehicles with high residual values ​​are more likely to yield a higher recovery value upon future disposal.

[0063] Vehicles can be broadly categorized into commercial and personal use. Personally used vehicles typically offer greater stability and a slower depreciation rate, while commercial vehicles, due to frequent use and higher wear and tear, depreciate more rapidly. Therefore, personally used vehicles carry relatively lower risk when used as collateral, while commercial vehicles require more careful evaluation.

[0064] Vehicle brand and model: Different brands and models of vehicles have varying resale value and market liquidity. Well-known brands and high-end models typically have higher market recognition and stronger resale value, while some niche brands or lower-end models have relatively weaker resale value. Therefore, vehicle brand and model are also taken into consideration during car loan assessments.

[0065] The purchase channel also affects the value and reliability of a vehicle as collateral. Vehicles purchased through formal channels typically have complete documentation and high transparency, while vehicles purchased through informal channels may have incomplete documentation, unclear ownership, and other issues, increasing the risk of improper disposal. Therefore, financial institutions tend to accept vehicles purchased through formal channels as collateral.

[0066] Vehicle mileage is an important indicator of a vehicle's condition. Vehicles with lower mileage generally have better condition and higher residual value, while vehicles with higher mileage may have higher maintenance costs and lower residual value. Therefore, vehicle mileage is one of the key factors to consider in auto loan assessments.

[0067] Vehicle insurance coverage is also a crucial factor in assessing vehicle risk. Comprehensive vehicle insurance not only protects against losses in the event of an accident but also enhances the overall safety of the vehicle. Therefore, vehicle insurance reports provided by insurance companies are one of the important bases for financial institutions to assess the risk of auto loans.

[0068] Therefore, by comprehensively assessing a vehicle's value, purpose, brand and model, purchase channel, mileage, and insurance status, vehicle-related characteristics can effectively identify the risks associated with using a vehicle as collateral. These characteristics not only reflect the vehicle's basic condition but also help financial institutions more accurately predict future repayment risks, thereby enabling them to take appropriate measures to manage and mitigate auto loan risks.

[0069] (4) Economic environment risk is also an important component of auto loan risk assessment, mainly involving the impact of macroeconomic factors and market conditions on the borrower's repayment ability. Specifically, these factors include the following aspects:

[0070] Macroeconomic conditions directly impact borrowers' employment opportunities, income levels, and overall financial situation. For example, slowing economic growth can lead to higher unemployment and reduced income, thereby increasing the risk of borrower default. Conversely, strong economic growth can improve borrowers' income levels and job stability, reducing the risk of default.

[0071] Interest rate changes directly impact borrowers' repayment burden. When interest rates rise, borrowers face an increased repayment load, potentially leading to some borrowers being unable to repay on time and increasing the risk of default. Conversely, falling interest rates can alleviate borrowers' repayment pressure and improve their ability to repay.

[0072] Inflation rates are a major factor affecting borrowers' financial health. High inflation erodes a borrower's real income, reducing their purchasing power and consequently impacting their ability to repay loans. High inflation also forces borrowers to pay more interest and principal when making repayments, increasing their financial burden. Conversely, low inflation helps maintain a borrower's purchasing power and income levels, reducing the risk of default.

[0073] Regional economic conditions also affect a borrower's repayment ability. Economically developed regions offer more job opportunities and higher income levels, leading to stronger repayment capacity among borrowers. Conversely, less developed regions have fewer job opportunities, lower income levels, weaker repayment capacity, and a higher risk of default among borrowers.

[0074] Changes in policies and regulations can also impact a borrower's repayment ability. For example, government tax and credit policies can directly affect a borrower's income and expenses. Furthermore, changes in regulatory policies, such as stricter credit checks and higher down payment requirements, can also affect a borrower's repayment ability.

[0075] Market competition also affects a borrower's repayment ability. In a highly competitive market environment, borrowers may face greater financial pressure, thus increasing the risk of default. Conversely, in a more stable market environment, a borrower's repayment ability is generally more stable.

[0076] Therefore, by comprehensively assessing macroeconomic conditions, interest rate changes, inflation rates, regional economic conditions, policy and regulatory changes, and market competition, economic environment risk can effectively identify borrowers' repayment risks under changing economic conditions.

[0077] (5) User personal characteristics are a key component of car loan risk assessment, mainly involving the borrower's personal information, credit history, repayment willingness and behavior, etc., in order to assess their reliability and risk level as a borrower.

[0078] Age and occupation: Research has found that a borrower's age and occupation have a significant impact on their repayment ability and stability. Young borrowers in the early stages of their careers may lack stable work experience and higher income levels, while older borrowers with stable jobs typically have higher incomes and better career prospects. Furthermore, whether a borrower has a stable occupation, such as a civil servant, teacher, or doctor, is also important, as these professions generally offer higher income stability and repayment ability.

[0079] Credit history is a crucial basis for assessing a borrower's creditworthiness. A good credit history indicates a borrower has a good repayment record and good credit standing, while a poor credit history suggests that the borrower may have had late payments or other credit problems. Financial institutions typically refer to a borrower's credit report to assess their credit score and repayment history.

[0080] Income and debt situation: A borrower's income level and debt situation are important factors in assessing their repayment ability. The higher the borrower's income and the lower their debt, the stronger their repayment ability and the lower their risk of default. Financial institutions will assess the stability of their income and the proportion of their debt by reviewing materials such as income statements and bank statements provided by the borrower.

[0081] A borrower's asset status is also an important factor in assessing their repayment ability. Possessing high-value fixed assets (such as real estate, stocks, and savings) can serve as additional repayment security. Financial institutions assess a borrower's asset status to understand their financial stability and potential sources of repayment.

[0082] Willingness and behavior to repay debts are also important indicators for assessing a borrower's credit risk. A borrower's good repayment habits, such as timely and early repayment, reflect their serious attitude towards debt. Furthermore, whether a borrower has a history of other loans or credit card use, and whether those records are positive, will also affect their credit score.

[0083] Educational background also reflects a borrower's repayment ability and creditworthiness to some extent. Borrowers with higher education generally have higher professional competence and repayment ability, while borrowers with lower levels of education may face greater financial pressure and repayment difficulties.

[0084] Family circumstances, such as the number of family members and marital status, also affect a borrower's repayment ability and stability. Borrowers from single-parent families or those with heavy family burdens may face greater financial pressure, while borrowers with fewer family members, stable marriages, and less financial burdens generally have a higher repayment ability.

[0085] Therefore, by comprehensively assessing a borrower's age and occupation, credit history, income and debt situation, asset status, repayment willingness and behavior, educational background and family situation, user personal characteristics can effectively identify a borrower's credit risk and repayment ability.

[0086] (6) Vehicle location information. This information can be used to obtain information such as vehicle driving trajectory, speed, and mileage. This information helps to more accurately grasp the borrower's vehicle usage, detect abnormal behavior in a timely manner, and thus improve the accuracy of risk prediction.

[0087] As an optional implementation, to inform the risk assessment model which dynamic characteristics are more important and which are relatively less important, the delinquency risk characteristics for the current repayment period include multiple delinquency sub-characteristics and preset weights assigned to each sub-characteristic. This can be understood as follows: during risk assessment, the borrower's delinquency risk characteristics for the current repayment period are subdivided into multiple specific sub-characteristics, and each sub-characteristic is assigned a pre-set weight. This design ensures that the impact of different characteristics on the overall risk can be reasonably quantified and reflected during the risk assessment process.

[0088] For example, the server can group various dynamic features according to different time levels, such as daily, weekly, monthly, and yearly changes; then, it can encode the non-numerical features in each group for further processing and calculation. When assigning weights, the server can assign weights to each sub-feature based on factors such as the frequency of periodic changes, degree of intent, policy sensitivity, impact, and economic indicators.

[0089] Regarding the frequency of periodic changes, the server can analyze the frequency of periodic changes of each dynamic feature and assign different weights based on the frequency. For example, if a feature changes every day, it will be given a higher weight; conversely, if a feature changes less frequently, it will be given a lower weight.

[0090] Regarding the level of intent, the server can assign different weights based on the impact of features on the level of intent to purchase, replace, or take out a loan within different time periods. For example, if users have a higher intent to take out a loan in February, then the relevant features for that month will be given a higher weight.

[0091] Regarding policy sensitivity, the server can analyze dynamic features that significantly impact policy sensitivity, such as the new energy vehicle replacement subsidy policy, and assign different weights based on the level of sensitivity. If a policy change has a significant impact on risk characteristics, then the features related to that policy will be given higher weights.

[0092] Regarding the degree of influence, the server can consider factors such as a country's vehicle management, production, and monetary policy for the current year, assigning different weights based on the degree of influence these factors have on dynamic features. For example, if a region's GDP grows rapidly, it will affect local loan demand and repayment ability, thus giving the region's features a higher weight.

[0093] For economic indicators, the server can weight the affected dynamic characteristics based on factors such as national and provincial GDP and per capita income to reflect the impact of the economic environment on risk characteristics. For example, if a province has a high per capita income, then the loan risk characteristics of that province will be assigned a lower weight.

[0094] Based on the above explanation of the borrower's risk characteristics, the following will continue to explain step S2 in Figure 1:

[0095] S2 uses a risk assessment model with a neural network architecture to process the characteristics of overdue risk and obtain the borrower's first risk score for the current repayment period.

[0096] In this embodiment, the risk assessment model based on the aforementioned neural network architecture can be a finely tuned large language model. Because this model has learned to understand and process complex textual information through extensive training data and can capture the correlations and potential patterns between different risk characteristics, when the borrower's risk characteristics are input into the risk assessment model in text form, the risk model first parses and understands the input text, identifying key risk characteristics and their weights. Next, the model uses its internal neural network structure to analyze how these characteristics collectively affect the overall risk through a series of calculations and inferences. Finally, the model outputs a comprehensive first risk score, which reflects the model's prediction of the borrower's delinquency risk within the current repayment period.

[0097] Therefore, in this way, the fine-tuned large language model can not only efficiently process diverse input data but also provide more accurate and reliable risk assessment results. Furthermore, through additional training data and fine-tuning, the model learns how to quickly adapt to and understand emerging information or changes. When faced with sudden events, the model can rapidly parse relevant text data, identifying key information and anomalies. For example, during risk assessment, if there are sudden major market fluctuations or news reports of significant events affecting a borrower's creditworthiness, the fine-tuned large language model can promptly capture this information and adjust the risk assessment results accordingly.

[0098] For example, the large language model could be Qwen-14B. This model is a deep learning system trained on a large scale, possessing powerful text understanding and generation capabilities. Therefore, by fine-tuning Qwen-14B, it can be used to predict the risk of borrower delinquency.

[0099] Based on the above embodiments explaining the method for obtaining the first risk score, the following will continue to explain step S3 in Figure 1:

[0100] S3 uses a pre-fitted autoregressive moving average model to process the delinquency risk characteristics and obtain the borrower's second risk score for the current repayment period.

[0101] It should be understood that the ARMA model, as a commonly used time series analysis tool, can capture and smooth the trend and seasonal components in data. In this application, the model first fits the collected historical repayment data of borrowers to determine the model parameters. Then, using these parameters, the model can process and predict the borrower's delinquency risk characteristics. To achieve this, the ARMA model can identify and quantify the time series characteristics of various risk factors, thereby calculating the borrower's secondary risk score within the current repayment period.

[0102] Compared to large language models, ARMA models rely on numerical inputs to perform calculations and analyses, and therefore cannot directly handle non-numerical data. For ARMA models to work effectively, the input risk sub-features must be converted into quantitative forms, i.e., transformed into numerical values ​​through some method. For example, some qualitative characteristics of borrowers (such as credit history and repayment behavior) may need to be quantified through scoring or coding. This quantified data can better represent the changing trends and patterns of risk characteristics, enabling ARMA models to capture the time-series properties of these characteristics for trend analysis and prediction.

[0103] In some embodiments, it is understood that when faced with sudden events, since the risk assessment model based on the neural network architecture can quickly parse the relevant text data and identify key information and anomalies, the aforementioned first risk score can characterize the impact of sudden information in the feature change information on the borrower's repayment risk in the current repayment period.

[0104] In some embodiments, since the autoregressive moving average model can identify and quantify the time-series characteristics of various risk factors and calculate the borrower's second risk score in the current repayment period, the second risk score can characterize the impact of non-sudden information in the feature change information on the borrower's repayment risk in the current repayment period.

[0105] Based on the above description of the method for obtaining the second risk score, the following continues the explanation of step S4 in Figure 1:

[0106] S4. Based on the first risk score and the second risk score, the borrower's overall risk score is obtained.

[0107] Thus, by combining the first risk score generated by the risk assessment model with the second risk score generated by the ARIMA model, the advantages of both models can be utilized to improve the overall accuracy of the prediction.

[0108] In this embodiment, both the first risk score and the second risk score are values ​​between 0 and 1, with higher values ​​indicating a higher risk of delinquency. Therefore, as an optional implementation, the first risk score and the second risk score can be weighted and summed using preset weights to obtain a comprehensive risk score.

[0109] Optionally, the borrower's comprehensive risk score is obtained based on the first risk score and the second risk score, including:

[0110] The borrower's overall risk score is calculated based on the first risk score, the second risk score, and the respective weights of the first and second risk scores.

[0111] Optionally, the weights of the first risk score and the second risk score can be preset weights, or they can be predicted in real time based on the weight model, which is not limited here.

[0112] In some implementations, when the difference between the first risk score and the second risk score is small, a preset weight can be used for calculation, so that when no sudden information occurs, the advantages of the two models can be combined to improve the overall prediction accuracy; while when the difference between the first risk score and the second risk score is large, the prediction weight obtained in real time from the weight model is used for calculation, so that when sudden information occurs, the overall prediction accuracy can be guaranteed.

[0113] Optionally, the weighting model can be a pre-trained machine learning model, such as a neural network or regression model, which can automatically learn and adjust the weights of each risk score based on optimized feature change information.

[0114] However, in practice, it has been found that when the difference between the first risk score and the second risk score is large, it means that the recognition results of the two models have diverged significantly. This is often caused by the ARIMA model's insensitivity to sudden information. Therefore, the preset weights cannot adapt to the above situation. As shown in Figure 2, this embodiment provides the following implementation method for step S4:

[0115] S4-1, obtain information on the changes in delinquency risk characteristics between the previous repayment period and the current repayment period.

[0116] Optionally, the overdue risk characteristics include: multiple risk sub-characteristics. The aforementioned information on the changes in overdue risk characteristics between the previous repayment period and the current repayment period includes:

[0117] Based on the risk sub-features in the overdue risk characteristics of the previous repayment period and the risk sub-features in the overdue risk characteristics of the current repayment period, calculate the feature change information.

[0118] Among them, the feature change information includes multiple change sub-information corresponding to multiple risk sub-features.

[0119] Optionally, for each risk sub-feature, if the change sub-information corresponding to the risk sub-feature meets the preset requirements, then the change sub-information corresponding to the risk sub-feature is determined to be sudden information; if the change sub-information corresponding to the risk sub-feature does not meet the preset requirements, then the change sub-information corresponding to the risk sub-feature is determined to be non-sudden information.

[0120] In this embodiment, the server can subtract each risk sub-feature from the overdue risk characteristics of the previous repayment period from each risk sub-feature of the current repayment period to obtain feature change information. If the change value of a certain risk sub-feature is large, it indicates that this change may be sudden information, reflecting new risk factors recently encountered by the borrower; conversely, if the change value is small, it may be non-sudden information, that is, normal fluctuations in the borrower's risk characteristics. In this way, sudden information and non-sudden information can be distinguished more accurately.

[0121] For example, as shown in Figure 3, the figure illustrates 5 risk sub-features from the previous repayment period and 5 risk sub-features from the current repayment period. The corresponding feature values ​​are shown in the table below:

[0122] As can be seen from the feature change information in Figure 3, risk sub-features A and B show increased changes over two adjacent repayment periods, while other risk sub-features show smaller changes. It should be understood that the above implementation is merely an example provided for ease of explanation regarding feature change information. The features and values ​​in the examples will not change with actual sub-feature changes.

[0123] Based on the above explanation of the feature change information, referring to Figure 2, after step S4-1, step S4 further includes:

[0124] S4-2, optimize the feature change information to obtain optimized feature change information.

[0125] The optimized feature change information enhances the sudden information within the feature change information while suppressing the non-sudden information. It should be understood that, as shown in Figure 3, the range of feature values ​​differs significantly for different risk sub-features; therefore, they cannot be directly used to measure the degree of change in an event. To address this, this embodiment proposes an optimization method that integrates the difference between the first and second risk scores with the feature change information to enhance the sudden information and suppress the non-sudden information. As shown in Figure 4, step S4-2 may include:

[0126] S4-2-2, based on the first risk score and the second risk score, obtain the divergence index between the risk assessment model and the autoregressive moving average model.

[0127] S4-2-3, use the divergence index to optimize the feature change information to obtain the optimized feature change information.

[0128] As shown in Figure 3, the feature change information includes multiple sub-information items. Therefore, in a specific implementation, for each sub-information item, the server can perform an exponentiation operation between the sub-information item and the divergence index to obtain the optimized sub-information item, where the reciprocal of the sub-information item is the exponent and the divergence index is the base. The corresponding mathematical expression is: y = x n

[0129] In the formula, y represents the optimized sub-information of change, x represents the divergence index, and n represents the reciprocal of the sub-information of change.

[0130] In this embodiment, the divergence index can be the ratio or difference between the first risk score and the second risk score. In other words, in some embodiments, the divergence index between the risk assessment model and the autoregressive moving average model can be determined based on the ratio or difference between the first risk score and the second risk score.

[0131] When the divergence index is the difference between the two values, it also ranges from 0 to 1. Through the above expression, if the sub-information of the change is less than 1, it will be further reduced; if the sub-information of the change is greater than 1, it will be further enhanced. Furthermore, constraining the value range of the optimized sub-information of the change to between 0 and 1 is beneficial for subsequent weight model processing.

[0132] This effectively addresses the issue of varying value ranges for different risk sub-features, ensuring that all feature change information is compared and processed on the same scale. Secondly, using the divergence index for power-law optimization allows for flexible amplification or reduction of the magnitude of change sub-information, thus more accurately distinguishing between sudden and non-sudden information. Specifically, for change sub-information less than 1, optimization further reduces its impact, while for change sub-information greater than 1, it further enhances its impact, enabling the model to more sensitively capture key change points.

[0133] The optimization method for feature change information based on the above implementation method is explained below. Referring to Figure 2 or Figure 4, after step S4-2, step S4 further includes:

[0134] S4-3 uses a pre-trained weight model to process the optimized feature change information to obtain the predicted weights for the first risk score and the second risk score.

[0135] S4-4: Based on the prediction weights of the first risk score and the second risk score, the comprehensive risk score is obtained.

[0136] Thus, the weighting model, by processing optimized feature change information, determines the importance of each risk score in the overall assessment. After training, the weighting model can more sensitively capture key changes and automatically adjust the weights of each risk score to more accurately reflect its predictive ability for future risks. Furthermore, this weighting model can be a pre-trained machine learning model, such as a neural network or regression model, which can automatically learn and adjust the weights of each risk score based on optimized feature change information. Specifically, the model can employ supervised learning methods, trained with a large amount of data, enabling it to identify which feature changes are more important for predicting the first and second risk scores. During training, the model continuously adjusts its internal parameters to minimize prediction error, ultimately resulting in a model capable of predicting the weights for each risk score.

[0137] As described in the above embodiments, when the difference between the first risk score and the second risk score is large, it means that the recognition results of the two models have diverged significantly. This is often caused by the ARIMA model's insensitivity to sudden information. Therefore, it is not always necessary to use the above weight model to adaptively generate the predicted weights for the first and second risk scores. Instead, it is necessary to rely on the divergence index between the first and second risk scores. Therefore, as shown in Figure 5, based on Figure 4, step S4-2 in Figure 4 further includes:

[0138] S4-2-1, Determine whether the divergence index is greater than the divergence threshold. If yes, proceed to step 4-2-2 in Figure 5; otherwise, proceed to sub-step S4-5 of step S4 in Figure 5.

[0139] S4-5, based on the preset weights of the first risk score and the second risk score, obtain the comprehensive risk score.

[0140] This can be understood as the server calculating a discrepancy index between the first and second risk scores and comparing it to a pre-set discrepancy threshold. If the discrepancy index does not exceed the threshold, the results of the two models are considered acceptable, and the combined risk score can be calculated directly using their respective pre-set weights. This not only avoids inaccurate weight allocation in high-discrepancy scenarios but also simplifies the calculation process in low-discrepancy scenarios, improving the overall system efficiency and robustness.

[0141] For the aforementioned risk assessment model and autoregressive moving average model, this embodiment proposes a novel model training method that can significantly improve the training efficiency. The training method for the risk assessment model will be explained in detail below using the risk assessment model as an example. Before proceeding, it should be understood that hyperparameters are crucial in model training, determining the model's learning process and final performance. Appropriate hyperparameter settings can control model complexity, prevent overfitting or underfitting, and thus ensure effective model learning; they also affect the model's convergence speed, and proper selection can accelerate the training process; furthermore, hyperparameters influence the selection and efficiency of optimization algorithms, thereby indirectly affecting the model's generalization ability.

[0142] In light of this, the server can also call a third-party large language model to generate the current hyperparameters for the network model to be trained; initialize the network model to be trained using the current hyperparameters; train the initialized network model to be trained using training samples to obtain candidate models trained by the network model to be trained and the training effect; continue to call the third-party large language model to process the current hyperparameters and training effect to obtain new hyperparameters; based on the new hyperparameters, return to the step of initializing the network model to be trained using the current hyperparameters, until the iteration stopping condition is met, and then select the model with the best training effect from multiple candidate models as the risk assessment model.

[0143] This can be understood as follows: in the above training method, the server uses the reasoning ability of a third-party large language model to generate hyperparameters for the network model to be trained, and adjusts the hyperparameters according to the feedback training effect, thereby continuously optimizing the hyperparameters set.

[0144] For example, assume the third-party large language model is Qwen72b and the network model to be trained is Qwen14b. In the early stages of training, a series of hyperparameter constraints are pre-set to optimize the model's training performance. First, the initial weight range is set to {initial weight / 500, initial weight*500}, and the learning rate is set to the interval [0.001, 0.1] with a step size of 0.001. The batch size is also set to the interval [16, 128] with a step size of 16. Finally, the number of hidden layer nodes is set according to the model complexity, ranging from [32, 256] with a step size of 32.

[0145] Based on the aforementioned constraints, a third-party large language model generates hyperparameters within these constraints. To achieve this, the server calls the third-party large language model Qwen72b to generate a set of initial hyperparameters for the network model Qwen14b to be trained. These hyperparameters are used to initialize the Qwen14b model. Next, the initialized Qwen14b model is trained using training samples to obtain a candidate model and its training performance. Subsequently, the server calls Qwen72b again, using the current hyperparameters and training performance as input, allowing Qwen72b to infer new hyperparameters based on the context of the call and the current hyperparameters and training performance. Based on the new hyperparameters, the server reinitializes the Qwen14b model and repeats the above training and hyperparameter tuning process until the iteration stopping condition is met (e.g., the loss value of the candidate model approaches stability). Throughout this process, the server utilizes Qwen72b's inference capabilities and the feedback training performance to adjust the hyperparameters, gradually optimizing the model. Finally, the model with the best training performance among multiple candidate models is selected as the final risk assessment model. In this way, the solution effectively utilizes the reasoning capabilities of Qwen72b to achieve efficient training and optimization of the Qwen14b model.

[0146] The same optimization method can be used to optimize the ARIMA model, and this embodiment will not elaborate on it further.

[0147] Based on the same inventive concept as the car loan risk prediction method provided in this embodiment, this embodiment also provides a car loan risk prediction device. This device includes at least one software functional module that can be stored in a memory or embedded in an electronic device. The processor in the electronic device executes the executable module stored in the memory. For example, the software functional modules and computer programs included in this device. Referring to Figure 6, functionally, this device may include:

[0148] Risk characteristics module 11 is used to obtain the borrower's overdue risk characteristics in the current repayment period;

[0149] Risk scoring module 12 is used to process the overdue risk characteristics through a risk assessment model with a neural network architecture to obtain the borrower's first risk score in the current repayment period;

[0150] The risk scoring module 12 is also used to process the overdue risk characteristics through a pre-fitted autoregressive moving average model to obtain the borrower's second risk score for the current repayment period.

[0151] The risk integration module 13 is used to obtain the borrower's comprehensive risk score based on the first risk score and the second risk score.

[0152] In this embodiment, the risk feature module 11 is used to implement step S1 in Figure 1, the risk scoring module 12 is used to implement steps S2 and S3 in Figure 1, and the risk synthesis module 13 is also used to implement step S4 in Figure 1. Therefore, for a detailed description of each of the above modules, please refer to the specific implementation method of the corresponding step, and this embodiment will not repeat it.

[0153] Since it shares the same concept as the aforementioned auto loan risk prediction method, the auto loan risk prediction device can also implement other steps or sub-steps of the method through the aforementioned module.

[0154] Optionally, the risk integration module 13 is also specifically used for:

[0155] Obtain information on the changes in delinquency risk characteristics between the previous repayment period and the current repayment period;

[0156] The feature change information is optimized to obtain the optimized feature change information. The optimized feature change information enhances the burst information in the feature change information while suppressing the non-burst information in the feature change information.

[0157] The optimized feature change information is processed by a pre-trained weight model to obtain the prediction weights for the first risk score and the second risk score, respectively.

[0158] The comprehensive risk score is obtained by calculating the prediction weights of the first risk score and the second risk score.

[0159] Optionally, the risk integration module 13 is also specifically used for:

[0160] Based on the first risk score and the second risk score, the divergence index between the risk assessment model and the autoregressive moving average model is obtained.

[0161] The feature change information is optimized by using the divergence index to obtain the optimized feature change information.

[0162] Optionally, the feature change information includes multiple sub-information pieces of change, and the optimized feature change information includes multiple optimized sub-information pieces of change; the risk integration module 13 is also specifically used for:

[0163] For each change sub-information, the change sub-information and the divergence index are exponentially operated to obtain the optimized change sub-information, where the reciprocal of the change sub-information is the exponent and the divergence index is the base.

[0164] Optionally, after obtaining the divergence index between the risk assessment model and the autoregressive moving average model based on the first risk score and the second risk score, the risk synthesis module 13 is further specifically used for:

[0165] Determine whether the divergence index is greater than the divergence threshold;

[0166] If the value exceeds the divergence threshold, then the step of optimizing the feature change information using the divergence index is executed to obtain the optimized feature change information.

[0167] Optionally, the delinquency risk characteristics of the current repayment period include multiple delinquency sub-characteristics and preset weights assigned to each delinquency sub-characteristic.

[0168] Optionally, the auto loan risk prediction device also includes a model training module, which is used for:

[0169] Call a third-party large language model to generate the current hyperparameters for the network model to be trained;

[0170] Initialize the network model to be trained using the current hyperparameters;

[0171] The initialized network model to be trained is trained using training samples to obtain the candidate model trained by the network model to be trained and the training effect.

[0172] Continue to call a third-party large language model to process the current hyperparameters and training results to obtain new hyperparameters;

[0173] Based on the new hyperparameters, return to the step of initializing the network model to be trained using the current hyperparameters until the iteration stopping condition is met, and then select the model with the best training performance from multiple candidate models as the risk assessment model.

[0174] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0175] It should also be understood that if the above embodiments are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0176] Therefore, this embodiment also provides a storage medium, which is a computer-readable storage medium. This storage medium stores a computer program, which, when executed by a processor, implements the car loan risk prediction method provided in this embodiment. The storage medium can be any medium capable of storing program code, such as a USB flash drive, external hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0177] This embodiment provides an electronic device for implementing a car loan risk prediction method. As shown in FIG7, the electronic device may include a processor 22 and a memory 21. The memory 21 stores a computer program, and the processor implements the car loan risk prediction method provided in this embodiment by reading and executing the computer program corresponding to the above-described embodiments in the memory 21.

[0178] Referring again to Figure 7, the electronic device also includes a communication unit 23. The memory 21, processor 22, and communication unit 23 are electrically connected to each other directly or indirectly via system bus 24 to realize data transmission or interaction.

[0179] The memory 21 can be an information recording device based on any electronic, magnetic, optical, or other physical principles, used to record execution instructions, data, etc. In some embodiments, the memory 21 can be, but is not limited to, volatile memory, non-volatile memory, memory drive, etc.

[0180] In some embodiments, the volatile memory may be random access memory (RAM); in some embodiments, the non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory, etc.; in some embodiments, the storage drive may be a disk drive, solid-state drive, any type of storage disk (such as optical disc, DVD, etc.), or similar storage media, or a combination thereof.

[0181] The communication unit 23 is used to send and receive data over a network. In some embodiments, the network may include a wired network, a wireless network, a fiber optic network, a telecommunications network, an intranet, the Internet, a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a metropolitan area network (MAN), a public switched telephone network (PSTN), a Bluetooth network, a ZigBee network, or a near field communication (NFC) network, or any combination thereof. In some embodiments, the network may include one or more network access points. For example, the network may include wired or wireless network access points, such as base stations and / or network switching nodes, through which one or more components of the service request processing system can connect to the network to exchange data and / or information.

[0182] The processor 22 may be an integrated circuit chip with signal processing capabilities, and may include one or more processing cores (e.g., a single-core processor or a multi-core processor). By way of example only, the processor described above may include a Central Processing Unit (CPU), an Application Specific Integrated Circuit (ASIC), an Application Specific Instruction-set Processor (ASIP), a Graphics Processing Unit (GPU), a Physics Processing Unit (PPU), a Digital Signal Processor (DSP), a Field Programmable Gate Array (FPGA), a Programmable Logic Device (PLD), a controller, a microcontroller unit, a Reduced Instruction Set Computing (RISC) computer, or a microprocessor, or any combination thereof.

[0183] It is understood that the structure shown in Figure 7 is for illustrative purposes only. Electronic devices may have more or fewer components than those shown in Figure 7, or may have different configurations. The components shown in Figure 7 may be implemented using hardware, software, or a combination thereof.

[0184] It should be understood that the apparatus and methods disclosed in the above embodiments can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0185] The above descriptions are merely various embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims. Industrial applicability

[0186] By adopting the above approach, the risk assessment model based on the neural network architecture can better capture the impact of sudden events, thereby making up for the insensitivity of the autoregressive moving average model to sudden events and improving the prediction accuracy of overdue risk.

Claims

1. A method for predicting auto loan risk, characterized in that, The method includes: Obtain the borrower's delinquency risk characteristics for the current repayment period; The overdue risk characteristics are processed by a risk assessment model based on a neural network architecture to obtain the borrower's first risk score in the current repayment period; The delinquency risk characteristics are processed by a pre-fitted autoregressive moving average model to obtain the borrower's second risk score in the current repayment period; The borrower's overall risk score is obtained based on the first risk score and the second risk score.

2. The auto loan risk prediction method according to claim 1, characterized in that, Based on the first risk score and the second risk score, the borrower's comprehensive risk score is obtained, including: The borrower's overall risk score is calculated based on the first risk score, the second risk score, and the respective weights of the first and second risk scores.

3. The auto loan risk prediction method according to claim 1 or 2, characterized in that, Based on the first risk score and the second risk score, the borrower's comprehensive risk score is obtained, including: Obtain information on the changes in overdue risk characteristics between the previous repayment period and the current repayment period; The feature change information is optimized to obtain optimized feature change information, wherein the optimized feature change information enhances the burst information in the feature change information while suppressing the non-burst information in the feature change information; The optimized feature change information is processed by a pre-trained weight model to obtain the predicted weights of the first risk score and the second risk score, respectively. The comprehensive risk score is obtained based on the prediction weights of the first risk score and the second risk score.

4. The auto loan risk prediction method according to claim 3, characterized in that, The feature change information is optimized to obtain optimized feature change information, including: Based on the first risk score and the second risk score, the divergence index between the risk assessment model and the autoregressive moving average model is obtained; The feature change information is optimized using the divergence index to obtain optimized feature change information.

5. The auto loan risk prediction method according to claim 3 or 4, characterized in that, The overdue risk characteristics include: multiple risk sub-characteristics; obtaining the characteristic change information between the overdue risk characteristics of the previous repayment period and the overdue risk characteristics of the current repayment period includes: Based on each risk sub-feature in the overdue risk characteristics of the previous repayment period and each risk sub-feature in the overdue risk characteristics of the current repayment period, the feature change information is calculated, and the feature change information includes multiple change sub-information corresponding to multiple risk sub-features.

6. The auto loan risk prediction method according to claim 5, characterized in that, For each risk sub-feature, if the change sub-information corresponding to the risk sub-feature meets the preset requirements, then the change sub-information corresponding to the risk sub-feature is determined to be sudden information; If the change sub-information corresponding to the risk sub-feature does not meet the preset requirements, then the change sub-information corresponding to the risk sub-feature is determined to be non-sudden information.

7. The auto loan risk prediction method according to claim 4, characterized in that, The step of obtaining the divergence index between the risk assessment model and the autoregressive moving average model based on the first risk score and the second risk score includes: Based on the ratio or difference between the first risk score and the second risk score, a divergence index is determined between the risk assessment model and the autoregressive moving average model.

8. The auto loan risk prediction method according to claim 4, characterized in that, The optimized feature change information includes multiple optimized change sub-information; The feature change information is optimized using the divergence index to obtain optimized feature change information, including: For each of the aforementioned change sub-information, the change sub-information and the divergence index are exponentially operated to obtain the optimized change sub-information, wherein the reciprocal of the change sub-information is the exponent and the divergence index is the base.

9. The auto loan risk prediction method according to claim 4 or 7, characterized in that, After obtaining the divergence index between the risk assessment model and the autoregressive moving average model based on the first risk score and the second risk score, the method further includes: Determine whether the divergence index is greater than the divergence threshold; If the value is greater than the divergence threshold, then the step of optimizing the feature change information using the divergence index to obtain optimized feature change information is performed.

10. The auto loan risk prediction method according to any one of claims 1-9, characterized in that, The delinquency risk characteristics of the current repayment period include multiple delinquency sub-characteristics and preset weights assigned to each of the delinquency sub-characteristics.

11. The auto loan risk prediction method according to any one of claims 1-10, characterized in that, The method also includes a training method for the risk assessment model, the training method comprising: Call a third-party large language model to generate the current hyperparameters for the network model to be trained; The current hyperparameters are used to initialize the network model to be trained; The initialized network model to be trained is trained using training samples to obtain candidate models trained by the network model to be trained and the training effect. The third-party large language model is then called to process the current hyperparameters and the training results to obtain new hyperparameters. Based on the new hyperparameters, return to the step of initializing the network model to be trained using the current hyperparameters until the iteration stopping condition is met, and then select the model with the best training effect from the multiple candidate models as the risk assessment model.

12. A car loan risk prediction device, characterized in that, The device includes: The risk characteristics module is used to obtain the borrower's delinquency risk characteristics in the current repayment period; The risk scoring module is used to process the overdue risk characteristics through a risk assessment model with a neural network architecture to obtain the borrower's first risk score in the current repayment period; The risk scoring module is also used to process the overdue risk characteristics through a pre-fitted autoregressive moving average model to obtain the borrower's second risk score in the current repayment period. The risk integration module is used to obtain the borrower's comprehensive risk score based on the first risk score and the second risk score.

13. A storage medium, characterized in that, The storage medium stores a computer program, which, when processed by a processor, implements the car loan risk prediction method according to any one of claims 1-11.

14. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing a computer program, which, when processed by the processor, implements the car loan risk prediction method according to any one of claims 1-11.

Images