Risk prediction method and device, electronic equipment, medium and program product

Abstract

This application provides a risk prediction method, apparatus, electronic device, medium, and program product, which can be applied to the fields of artificial intelligence technology and fintech. The method includes: acquiring financial characteristic data of the object to be evaluated; inputting the financial characteristic data into multiple pre-constructed local Gaussian process regression models, outputting multiple default prediction values; obtaining a risk bifurcation diagram based on the multiple local Gaussian process regression models, wherein the risk bifurcation diagram is used to identify input variable thresholds that cause abrupt changes in the default prediction values ​​under the condition of disturbance of the financial characteristic data; and performing risk prediction of the object to be evaluated based on the multiple default prediction values ​​and the risk bifurcation diagram.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence technology and fintech, and more specifically to a risk prediction method, apparatus, device, medium, and program product. Background Technology

[0002] As a vital foundation for national economic and social development, micro and small enterprises play an irreplaceable role in promoting employment, increasing national income, and stimulating market vitality. However, micro and small enterprise lending faces challenges such as a significantly higher non-performing loan rate than the average for corporate loans and a mismatch between market contribution and loan availability.

[0003] Currently, assessments of loan default risk for micro and small enterprises (MSEs) are mostly based on traditional models with linear or smoothed nonlinear assumptions. These models cannot capture the abrupt changes that cause a surge in loan default rates when economic conditions fall below a threshold. Furthermore, traditional models neglect the feedback effect of banks' risk mitigation actions on MSE default decisions and ignore the systemic phase transitions that adjustments to certain factors may trigger. These shortcomings cause traditional models to fail in providing early warnings of sudden default waves and make it difficult to offer guidance for default rate control. Summary of the Invention

[0004] In view of the above problems, this application provides risk prediction methods, apparatus, equipment, media and program products.

[0005] According to a first aspect of this application, a risk prediction method is provided, the method comprising: acquiring financial characteristic data of an object to be evaluated; inputting the financial characteristic data into multiple pre-constructed local Gaussian process regression models respectively, and outputting multiple default prediction values; obtaining a risk bifurcation diagram based on the multiple local Gaussian process regression models, the risk bifurcation diagram being used to identify input variable thresholds that cause abrupt changes in the default prediction values ​​under the condition of disturbance of the financial characteristic data; and performing risk prediction of the object to be evaluated based on the multiple default prediction values ​​and the risk bifurcation diagram.

[0006] According to an embodiment of this application, obtaining a risk bifurcation diagram based on the plurality of local Gaussian process regression models includes: constructing an input variable space based on multiple input variables in the financial feature data; generating a corresponding bifurcation curve for each local Gaussian process regression model and the input variable space; and constructing the risk bifurcation diagram based on the multiple bifurcation curves corresponding to the plurality of local Gaussian process regression models.

[0007] According to an embodiment of this application, the plurality of input variables includes at least a first input variable and a second input variable. The step of generating a corresponding bifurcation curve for each local Gaussian process regression model and the input variable space includes: determining a target point where the partial derivative of the default prediction value of the local Gaussian process regression model with respect to the direction of the second input variable is zero, under the condition that the first input variable is set to different fixed values; and connecting the target point continuously in the input variable space to generate the corresponding bifurcation curve.

[0008] According to an embodiment of this application, constructing the risk bifurcation diagram based on the multiple bifurcation curves corresponding to the multiple local Gaussian process regression models includes: splicing the multiple bifurcation curves in the input variable space according to numerical continuity to form a set of continuous curves; and constructing the risk bifurcation diagram based on the set of continuous curves.

[0009] According to an embodiment of this application, the step of splicing the multiple bifurcation curves in the input variable space according to numerical continuity includes: based on the last target point of the current bifurcation curve, performing linear extrapolation in the input variable space using a pseudo-arc length numerical continuity method to predict the initial target point position of the target bifurcation curve; under the condition that the first input variable of the initial target point position is set to a fixed value, determining the prediction target point where the partial derivative of the corresponding local Gaussian process regression model with respect to the default prediction value in the direction of the second input variable is zero; and sequentially connecting the prediction target point and the target point of the current bifurcation curve to splice the current bifurcation curve and the target bifurcation curve.

[0010] According to an embodiment of this application, the step of predicting the risk of the object to be evaluated based on the plurality of default prediction values ​​and the risk bifurcation diagram includes: forming a plurality of predicted state points in the input variable space based on the plurality of default prediction values; and predicting the risk of the object to be evaluated based on the determination of whether the plurality of predicted state points fall into the mutation boundary region of the risk bifurcation diagram.

[0011] According to an embodiment of this application, the method further includes: in response to the prediction uncertainty of the current local Gaussian process regression model exceeding a preset threshold, taking the current target point of the current bifurcation curve as a stopping point; and establishing a target local Gaussian process regression model based on the position of the stopping point in the input variable space, wherein the target local Gaussian process regression model is used to generate the target bifurcation curve.

[0012] A second aspect of this application provides a risk prediction device, comprising: a data acquisition module for acquiring financial characteristic data of an object to be evaluated; a default prediction value acquisition module for inputting the financial characteristic data into multiple pre-constructed local Gaussian process regression models and outputting multiple default prediction values; a risk bifurcation diagram acquisition module for obtaining a risk bifurcation diagram based on the multiple local Gaussian process regression models, wherein the risk bifurcation diagram is used to identify input variable thresholds that cause abrupt changes in the default prediction values ​​under the condition of disturbance of the financial characteristic data; and a risk prediction module for predicting the risk of the object to be evaluated based on the multiple default prediction values ​​and the risk bifurcation diagram.

[0013] According to an embodiment of this application, the risk bifurcation diagram acquisition module can also be used to construct an input variable space based on multiple input variables in the financial feature data; generate a corresponding bifurcation curve for each local Gaussian process regression model and the input variable space; and construct the risk bifurcation diagram based on multiple bifurcation curves corresponding to the multiple local Gaussian process regression models.

[0014] According to an embodiment of this application, the risk bifurcation diagram acquisition module can also be used to determine the target point where the partial derivative of the default prediction value of the local Gaussian process regression model with respect to the direction of the second input variable is zero, under the condition that the first input variable is set to different fixed values; and to connect the target point continuously in the input variable space to generate the corresponding bifurcation curve.

[0015] According to embodiments of this application, the risk bifurcation diagram acquisition module can also be used to splice the multiple bifurcation curves in the input variable space according to numerical continuity to form a continuous curve set; and

[0016] Based on the set of continuous curves, the risk bifurcation diagram is constructed.

[0017] According to an embodiment of this application, the risk bifurcation diagram acquisition module can also be used to predict the initial target point position of the target bifurcation curve based on the final target point of the current bifurcation curve by performing linear extrapolation in the input variable space using the pseudo-arc length numerical continuity method; under the condition that the first input variable of the initial target point position is set to a fixed value, determine the prediction target point where the partial derivative of the default prediction value of the corresponding local Gaussian process regression model with respect to the direction of the second input variable is zero; and connect the prediction target point and the target point of the current bifurcation curve in sequence to splice the current bifurcation curve and the target bifurcation curve.

[0018] According to an embodiment of this application, the risk bifurcation diagram acquisition module can also be used to, in response to the prediction uncertainty of the current local Gaussian process regression model exceeding a preset threshold, take the current target point of the current bifurcation curve as a stopping point; based on the position of the stopping point in the input variable space, establish a target local Gaussian process regression model, which is used to generate the target bifurcation curve.

[0019] According to an embodiment of this application, the risk prediction module can also be used to respond to the fact that the prediction uncertainty of the current local Gaussian process regression model exceeds a preset threshold, and to take the current target point of the current bifurcation curve as a stop point; based on the position of the stop point in the input variable space, to establish a target local Gaussian process regression model, which is used to generate the target bifurcation curve.

[0020] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0021] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0022] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0023] According to embodiments of this application, by constructing multiple local Gaussian process regression models, the characteristics of nonlinear mutation behavior in financial risk systems can be accurately fitted, improving the predictive ability for sudden default events. Using local modeling instead of a global model significantly reduces the computational burden in high-dimensional space and improves resource utilization efficiency during training and prediction. Simultaneously, the constructed risk bifurcation diagram intuitively displays the critical relationship between input variables and default risk, providing users with a visualized and structured interpretation of risk, enhancing the transparency and decision-making support value of the model output, and helping users quickly understand and take targeted measures. Attached Figure Description

[0024] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0025] Figure 1 The illustrations depict application scenarios of risk prediction methods, apparatus, devices, media, and program products according to embodiments of this application.

[0026] Figure 2 A flowchart illustrating a risk prediction method according to an embodiment of this application is shown schematically.

[0027] Figure 3 The flowchart illustrates a method for obtaining a risk bifurcation plot based on a multiple local Gaussian process regression model according to some exemplary embodiments of this application.

[0028] Figure 4 The illustration shows a schematic diagram of a method for risk prediction of an object to be evaluated according to some exemplary embodiments of this application;

[0029] Figure 5 A schematic diagram illustrating the structure of a risk prediction device according to an embodiment of this application is shown; and

[0030] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a risk prediction method according to an embodiment of this application. Detailed Implementation

[0031] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0032] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0033] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0034] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0035] First, the technical terms used in this article are explained and clarified as follows.

[0036] Local Gaussian process regression models are Gaussian process regression models constructed based on a limited number of samples within a specific local region of the input variables. They are used to fit and predict data relationships within that local region. Compared to global modeling, local Gaussian process regression models offer greater flexibility and fitting accuracy, better adapting to regions in the input variable space where the function's form changes drastically. They are particularly suitable for handling complex systems with local mutations or multiple solution structures.

[0037] The input variable space refers to the multidimensional data space spanned by multiple financial characteristic variables of the object to be evaluated (i.e., model inputs). In the embodiments of this application, the input variable space constitutes the domain of the Gaussian process regression model, used to characterize all possible combinations of input variables. This space provides a structured modeling foundation for the risk prediction model and serves as the geometric construction region for bifurcation curves and risk bifurcation diagrams.

[0038] The pseudo-arc length numerical continuity method is a numerical computation method used to trace the solution path of implicit curves in parameter space. Its basic principle is to estimate the position of the next target point by linear extrapolation at a point on the curve and combining the direction of the line connecting that point and the previous solution point. Then, by iteratively solving for the true solution points that satisfy specific constraints (such as the derivative being zero), the entire curve is gradually extended.

[0039] As a vital component of the market, micro and small enterprises (MSEs) play an irreplaceable role in absorbing social employment, promoting local economic development, and enhancing the resilience of the industrial chain. Currently, the financial system's ability to serve MSEs is gradually increasing, and policy support for MSE financing is continuously being strengthened. However, in terms of actual loan implementation, MSE loans still face the following major problems: First, MSEs generally exhibit characteristics such as small scale, poor operational stability, and weak risk resistance, making them more prone to cash flow disruptions, loan defaults, or delinquencies during economic cycles or policy adjustments. Second, due to factors such as information asymmetry, lack of collateral, and an imperfect credit rating system, assessing the loan risks of MSEs is difficult, resulting in a consistently higher non-performing loan rate compared to other types of enterprises.

[0040] Existing models for loan default risk in micro and small enterprises primarily rely on statistical analysis of variables such as historical repayment behavior, corporate financial indicators, and industry classification. These methods often assume a linear or smooth nonlinear relationship between various influencing factors and default outcomes, using classification algorithms such as logistic regression, support vector machines, and gradient boosting trees to predict default probabilities. These models are typically built on the assumption of a stable economic environment and cannot effectively identify group default surges caused by certain variables falling below a critical point.

[0041] Furthermore, in actual lending operations, banks, as credit providers, directly impact loan costs, loan acquisition difficulty, and borrower repayment pressure through their risk management behaviors (such as adjusting risk provision ratios, implementing capital buffer policies, controlling total credit lines, and interest rate pricing strategies). Currently, most default risk models focus only on the behavioral characteristics of borrowing companies, neglecting the adjustments banks make after perceiving changes in credit market risk and their feedback effects on default behavior. Since bank actions can amplify market risk—for example, by withdrawing loans or raising interest rates when anticipated risk increases—they can compress corporate financing space and indirectly trigger widespread defaults. Traditional models lack the capacity to model such feedback loops, resulting in insufficient description of the evolution of systemic risk.

[0042] Furthermore, existing technologies fail to adequately consider the default mechanisms of systemic phase transitions. In the financial system, small and micro enterprises form interconnected networks through various means such as supply chains, geographical regions, or industries. When the default risk at a local node increases, it may spread throughout the entire system through network effects, triggering a large-scale wave of defaults. Traditional modeling methods based on independent default events fail to effectively integrate the contagious connections between enterprises, making it difficult to identify and intervene in potential systemic risks in advance.

[0043] Based on this, embodiments of this application provide a risk prediction method, the method comprising: initiating data requests to a server using a first device and a second device to obtain static rendering data; performing a first risk prediction on the first device and the second device based on the static rendering data; initiating data requests to the server using the first device at preset time intervals to obtain dynamic rendering data; and distributing the dynamic rendering data to corresponding second devices according to predetermined identifiers; and performing a second risk prediction on the second device based on the dynamic rendering data in response to the second device receiving the dynamic rendering data with the corresponding identifiers. According to embodiments of this application, by uniformly obtaining dynamic rendering data through the first device and distributing it to corresponding second devices according to predetermined identifiers, the problem of multiple terminals frequently requesting the same data from the server is avoided, thereby reducing the occupation of network bandwidth and server interface resources; by separating the processing of static rendering data and dynamic rendering data, rapid initial loading of the page and high-frequency real-time updates of subsequent data can be achieved, improving overall rendering performance and user experience; at the same time, the data distribution mechanism supports different second devices to accurately receive the required data according to their own identifiers, ensuring the consistency and relevance of page display, avoiding data redundancy and display inconsistencies between different terminals, and possessing high flexibility and scalability.

[0044] It should be noted that the risk prediction methods, apparatus, devices, media, and program products defined in this application can be used in the fields of artificial intelligence technology and fintech, and can also be used in a variety of other fields besides artificial intelligence technology and fintech. The application fields of the risk prediction methods, apparatus, devices, media, and program products provided in the embodiments of this application are not limited.

[0045] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0046] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this application all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results. If the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests, or economic, health, and credit status through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0047] Figure 1 The illustrations depict application scenarios of risk prediction methods, apparatus, devices, media, and program products according to embodiments of this application.

[0048] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0049] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0050] In the embodiments of this application, the first terminal device 101 can be an example of the first device, and the second terminal device 102 and / or the third terminal device 103 can be an example of at least one second device. The first device and the second device can communicate collaboratively through an internal client mechanism to implement the data distribution and rendering logic described in the risk prediction method.

[0051] In some embodiments, the first device and at least one second device may be different display modules, windows or screens on the same computing terminal (such as a host), or multiple physical devices that work together through a network, such as different client instances deployed on a desktop computer, tablet terminal or mobile device respectively.

[0052] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smart mobile terminals, tablet computers, laptop computers, and desktop computers.

[0053] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0054] It should be noted that the risk prediction method provided in this application embodiment can generally be executed by server 105. Correspondingly, the risk prediction device provided in this application embodiment can generally be located in server 105. The risk prediction method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the risk prediction device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0055] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0056] The following will be based on Figure 1 The described scene, through Figures 2-4 The risk prediction method of the disclosed embodiments is described in detail.

[0057] Figure 2 A flowchart illustrating a risk prediction method according to an embodiment of this application is shown.

[0058] like Figure 2 As shown, the risk prediction method 200 of this embodiment includes operations S210 to S240.

[0059] In operation S210, financial characteristic data of the object to be evaluated is obtained.

[0060] In the embodiments of this application, the objects to be evaluated may include different types of entities such as corporate customers, individual business owners, and natural person borrowers.

[0061] For example, for corporate clients, financial characteristic data may include key indicators from the balance sheet, cash flow statement, and profit and loss statement (such as total assets, total liabilities, net profit, and cash flow from operating activities), as well as financial ratio indicators closely related to the stability of the company's operations (such as current ratio, quick ratio, debt-to-equity ratio, and interest coverage ratio). Supplementary characteristics such as external rating information, default records in credit reports, and guarantee information can also be obtained.

[0062] In the embodiments of this application, feature data can be collected from multiple dimensions to improve the assessment coverage and adaptability. For example, micro and small enterprises generally suffer from incomplete or unaudited financial information. Therefore, the system can supplementarily collect their bank statements, tax declaration data, e-commerce operating indicators, online store ratings, and user reviews as proxy variables to construct a multi-dimensional representation of financial features. For example, for a micro and small clothing enterprise operating online sales, the system can collect its monthly turnover over the past year, monthly active days on the e-commerce platform, return rate, trends in major sales channels, as well as loan history, current liabilities, and query frequency.

[0063] Furthermore, the collection of financial characteristic data can also support time series structures to reflect the dynamic evolution trend of an object. For example, for a medium-sized manufacturing enterprise, the system can not only obtain its latest financial statements, but also collect time series of key financial indicators for 12 consecutive months (such as sales revenue, accounts receivable turnover, fixed asset investment, etc.), thereby constructing more forward-looking risk assessment inputs.

[0064] Regarding data sources, financial feature data can originate from data interfaces uploaded by enterprises themselves or those with data sharing agreements signed with financial institutions. It can also be automatically collected and structured through data connections established with supply chain finance platforms and enterprise credit reporting service providers. To ensure data integrity and consistency, the system can be configured with a feature missing tolerance mechanism, allowing for the handling of some missing items through interpolation, mean imputation, or model prediction.

[0065] In operation S220, the financial characteristic data is input into multiple pre-built local Gaussian process regression models, and multiple default prediction values ​​are output.

[0066] In the embodiments of this application, the construction of a local Gaussian process regression model may include segmented modeling of the financial feature space, that is, dividing the entire input space of the financial feature data into several local subspaces, and training a regression model with Gaussian process prior in each subspace separately, thereby improving the model's sensitivity and fitting ability to nonlinear structures and local mutations.

[0067] Specifically, clustering algorithms can be used to subdivide historical financial feature data into subspaces, with each subclass representing a region of approximate features. Local Gaussian process models are then trained based on data from each subclass. For new objects to be evaluated in practical applications, their financial feature vectors can be assigned to the corresponding local models using nearest neighbors or kernel similarity functions, or simultaneously fed into all local models to generate multiple parallel outputs of default probability estimates. These predictions can represent the object's default tendency from different local risk perspectives, thus providing a more detailed risk profile.

[0068] According to embodiments of this application, the system can fully leverage the advantages of local modeling when small and micro enterprises are the subjects of evaluation. For example, due to the significant heterogeneity of small and micro enterprises in terms of the completeness of financial data and the stability of their business models, it is difficult to accurately cover their diverse characteristic structures using a single global model. Therefore, by employing multiple local Gaussian process models, grouping and modeling can be done according to their business characteristics. For example, separate models can be built for e-commerce, catering, and manufacturing small and micro enterprises to capture the mapping relationship between their typical characteristics and default patterns. In some embodiments, the system can perform preliminary clustering based on indicators such as the sales scale, transaction frequency, and cash flow volatility of small and micro enterprises to form multiple local risk clusters, and train a set of Gaussian process models for each cluster. When a new subject of evaluation enters the evaluation process, its financial characteristic terms can be input into each cluster model, and multiple sets of corresponding default prediction values ​​can be output.

[0069] To enhance model adaptability, adaptive kernel functions or multi-kernel fusion techniques can be introduced, enabling each local Gaussian process model to automatically adjust its weight allocation for the sensitivity of specific variables. For example, for small and micro manufacturing enterprises, equipment investment levels and accounts receivable cycles may be sensitive variables, while for internet retail enterprises, gross profit margin and user retention rate may be more valuable as risk indicators. During training, local models can automatically learn the importance of these variables in their local space through Bayesian inference mechanisms, thereby improving prediction accuracy.

[0070] In operation S230, a risk bifurcation diagram is obtained based on the multiple local Gaussian process regression models. The risk bifurcation diagram is used to identify the input variable threshold that causes the default prediction value to undergo abrupt behavior under the perturbation of the financial characteristic data.

[0071] In the embodiments of this application, the risk bifurcation diagram can be used to visually demonstrate whether the default prediction value will undergo a significant jump when financial characteristic variables change within a certain perturbation range, thereby identifying the key variables that trigger drastic fluctuations in default probability and their corresponding threshold ranges. In practical applications, the risk bifurcation diagram not only helps enhance the interpretability of risk assessment results but also assists financial institutions in identifying potential risk-sensitive intervals, guiding the setting of credit boundaries, early warning, and intervention strategies.

[0072] Specifically, the system can perform local perturbation processing on each input variable. That is, while keeping the values ​​of other variables fixed, the value of the target variable is gradually increased or decreased, and the perturbed input vector is then input into each local Gaussian process regression model. The trajectory of the output default prediction value as the variable perturbs is recorded. By analyzing the trend of the prediction curve, if a steep change or discontinuous jump occurs near a specific value of a variable, that value can be identified as the risk threshold of that variable. The system further summarizes all significant change points in the input space to form a risk bifurcation diagram under high-dimensional feature perturbation.

[0073] For example, under normal circumstances, a medium-sized manufacturing company has a debt-to-equity ratio of 70%. When constructing a risk bifurcation diagram, the system finds that when the debt-to-equity ratio rises from 70% to 75%, the predicted default value does not change significantly. However, when it rises further to above 77%, the predicted value shows a step increase, indicating that 77% is the risk threshold for the company under its current condition. Similarly, the system can also identify sensitive intervals for other variables such as cash flow coverage ratio, revenue growth rate, or operating cycle, thereby forming a multivariate joint risk bifurcation diagram for a specific company.

[0074] In the embodiments of this application, since micro and small enterprises often have small data sample sizes and high volatility, traditional models struggle to accurately grasp their critical risk points. However, by utilizing the Bayesian modeling characteristics of local Gaussian processes, prediction uncertainty can be fully expressed, and perturbation analysis can quantify their response to input disturbances. For example, a micro and small online retail enterprise has a predicted default probability of 20% when its gross profit margin is 12%, which rises to 35% when the gross profit margin drops to 9%, and jumps to 65% when it drops to 7%. Based on this, the system determines that "gross profit margin less than 8%" is a highly sensitive threshold segment under the current business conditions, and suggests that the credit granting model should weight the criticality of this factor when assessing such enterprises.

[0075] Furthermore, risk bifurcation diagrams can also serve as strategy simulation tools, supporting reverse-engineering operational adjustment suggestions. For example, when a company is currently in a state of high default prediction, the system can use bifurcation diagram analysis to discover that adjusting several key variables (such as shortening the operating cycle and increasing net profit margin) to above a certain threshold can significantly reduce its default probability, thereby providing data support for risk mitigation and guidance mechanisms. This mechanism is also applicable to compliance scenarios such as bank credit condition design and guarantee scheme adjustments.

[0076] In operation S240, risk prediction of the object to be evaluated is performed based on the multiple default prediction values ​​and the risk bifurcation diagram.

[0077] In the embodiments of this application, multiple default prediction values ​​can be weighted and fused, with the weights automatically adjusted based on the applicability, prediction confidence, or historical performance of each local model. The fusion process not only outputs a comprehensive default probability value but also incorporates the distribution characteristics of the prediction values ​​(such as mean, variance, and skewness) to help identify the stability of the prediction results. For example, for a specific enterprise, its default prediction values ​​under multiple local Gaussian process regression models are 0.15, 0.20, 0.18, 0.80, and 0.22, respectively. The system not only outputs the average value as a comprehensive score but also identifies an abnormal sensitivity of the Nth model to the prediction results, indicating a potential high-risk branch requiring further investigation.

[0078] Risk bifurcation diagrams can provide crucial support for decision-making. The system can combine key variables identified in the bifurcation diagram with their mutation thresholds to perform a sensitivity scan of the current financial status of the entity being assessed, determining whether it is approaching a risk jump zone or has already fallen into a high-risk threshold range. For example, when assessing a type of small and micro manufacturing enterprise, if it is found that its debt-to-equity ratio is approaching the high-risk threshold in the historical bifurcation diagram, while its overall default probability remains in the median range, the system can further weight the sensitivity indicators of this characteristic, improving the overall risk rating and thus achieving dynamic correction and enhanced interpretation.

[0079] In the embodiments of this application, the prediction results can also be classified and output according to risk levels based on the needs of scenario-based strategy execution. For example, in the scenario of bank credit approval, the overall default probability can be divided into three intervals: "low risk (<10%)", "medium risk (10%-30%)" and "high risk (>30%)". Based on the bifurcation diagram, it can be determined whether there are "highly sensitive variable hit" or "structural instability" labels, which can be used to indicate whether further manual verification, supplementary materials or adjustment of credit structure are needed.

[0080] Furthermore, the system can dynamically generate strategy recommendations or business response plans based on risk prediction results. For example, when the prediction results show that the target is in a boundary-sensitive range, the system can recommend a conservative credit limit setting, additional collateral conditions, or requirements to increase the transparency of cash flow; if the target is currently in a safe range, but bifurcation analysis shows that its core variables have a future trend of change, a "mid-term risk monitoring plan" can be generated, setting a dynamic review cycle for it.

[0081] According to the embodiments of this application, by combining multiple local Gaussian process regression models with risk bifurcation diagrams, fine-grained modeling and dynamic threshold identification of the default risk of the object to be assessed are achieved. This not only improves the accuracy and stability of the prediction results, but also enhances the model's sensitivity and interpretability to financial feature disturbances. In particular, when dealing with entities with strong data heterogeneity and complex risk behaviors, such as micro and small enterprises, it can effectively identify key variables and their critical points that cause sudden changes in default probability, thereby providing more targeted and forward-looking intelligent decision support for credit approval, risk warning and strategy formulation.

[0082] The risk prediction method of this application will be specifically described below by way of preferred embodiments.

[0083] In the embodiments of this application, the process of obtaining a risk bifurcation diagram based on multiple local Gaussian process regression models can model and visualize the nonlinear response relationship between input variable disturbances and default probability prediction results, thereby identifying the risk-sensitive intervals and mutation thresholds of key variables.

[0084] Figure 3 The flowchart illustrates a method for obtaining a risk bifurcation plot based on multiple local Gaussian process regression models according to some exemplary embodiments of this application.

[0085] like Figure 3 As shown, the method for obtaining the risk bifurcation diagram based on multiple local Gaussian process regression models includes operations S310 to S330.

[0086] In operation S310, an input variable space is constructed based on multiple input variables in the financial feature data.

[0087] Specifically, an input variable space can be constructed first based on multiple input variables from financial feature data. This input variable space can be a multi-dimensional, continuous feature combination space, encompassing corporate financial indicators, operational behavior indicators, and external signal indicators. The system can then define the perturbation range and step size for each variable to ensure coverage of the entire reasonable value range and sufficient resolution to detect potential abrupt changes. The construction of the input variable space can also be based on the statistical distribution of historical data to determine the combination of principal variables, and principal component analysis or factor analysis methods can be used for dimensionality compression, improving the operability and interpretability of the bifurcation diagram construction.

[0088] In operation S320, for each local Gaussian process regression model and the input variable space, a corresponding bifurcation curve is generated.

[0089] In the embodiments of this application, for each local Gaussian process regression model, the system can, under the condition of fixing the values ​​of other variables, traverse or sample a single input variable within its perturbation interval, record the default prediction value output corresponding to each perturbation point, and then draw the response curve between variable perturbation and prediction output, which is called the bifurcation curve under the model. This curve reflects the sensitivity of the model to changes in the input variable.

[0090] For example, the multiple input variables may include a first input variable and a second input variable, and the input variable space can constitute a two-dimensional perturbation space as the input variable space. Optionally, any two dimensions of the multiple input variables, such as the first input variable and the second input variable, can be selected to construct the two-dimensional perturbation space.

[0091] In this embodiment, the differentiability of the local Gaussian process model can be utilized to perform a first-order partial derivative operation on the second input variable based on the model's prediction function. This allows the system to assess whether a small perturbation in the second input variable, given a fixed value for the first input variable, will cause a significant change in the default prediction value. If the derivative is zero, it indicates that near that point, the change in the second input variable has a potentially sensitive inflection point significance for the model output to reach an extreme value or inflection point. The system can define such points with zero partial derivatives as target points on the perturbation path.

[0092] By repeating the above process under different values ​​of the first input variable, the system can obtain a series of target points. These target points can be regarded as the trajectory or extreme value line of the risk prediction output in the input variable space. The system can further connect all target points sequentially in the input variable space to form a bifurcation curve with structural turning point significance. This bifurcation curve is used to reveal the boundary relationship of input variables that triggers abrupt changes in default probability under a specific combination of variables, and has strong risk interpretation value.

[0093] For example, when assessing the risk of a certain type of asset-heavy manufacturing enterprise, the system can select the debt-to-asset ratio as the first input variable and the net profit margin as the second input variable. Under a series of fixed debt-to-asset ratio levels (60%, 65%, 70%, 75%), the system calculates the partial derivative of the model in the direction of net profit margin, identifies the net profit margin values ​​where the derivative is zero, and records the coordinates of these points in the input variable space. The curve formed by connecting these points is a bifurcation curve representing a risk mutation structure in the "debt-to-asset ratio - net profit margin" space. This bifurcation curve can indicate under what financial structure combinations the predicted default risk of the enterprise may fluctuate drastically, thus providing data support for subsequent credit threshold setting and risk intervention trigger point identification.

[0094] According to the embodiments of this application, the bifurcation curve constructed based on the target point with a partial derivative of zero not only has a clear mathematical boundary meaning, but can also be extended to high-dimensional variable analysis through scanning of multidimensional combination space, thereby constructing a risk mutation structure map under the interaction between complex input variables, enhancing the risk assessment system's ability to identify and interpret complex input disturbance structures.

[0095] In operation S330, the risk bifurcation diagram is constructed based on the multiple bifurcation curves corresponding to the multiple local Gaussian process regression models.

[0096] In the embodiments of this application, the bifurcation curves generated by all local Gaussian process models in each input variable dimension can be summarized to construct a complete risk bifurcation diagram. The risk bifurcation diagram can be displayed using high-dimensional visualization methods (such as joint presentation of multiple subgraphs, heatmap mapping, 3D projection, etc.), or it can be converted into structured indicators, such as the "risk mutation index," "change initiation point," and "high-risk window range" for each variable, for subsequent use by the evaluation system.

[0097] In the embodiments of this application, the process of constructing a risk bifurcation diagram based on multiple bifurcation curves corresponding to multiple local Gaussian process regression models may include the following steps: First, in operation S320, the system generates one or more bifurcation curves based on each local Gaussian process regression model and the input variable space. These bifurcation curves represent the boundary paths where the default prediction value undergoes abrupt changes under the condition of input variable perturbation. The feature space locations of each bifurcation curve may be different, but they have continuity in numerical distribution and reflect the response characteristics of the model to the same combination of input variables.

[0098] Furthermore, each bifurcation curve can be projected into a unified input variable space to identify the spatial proximity of the curves and logically concatenate them according to numerical continuity. Specifically, the system can determine whether adjacent bifurcation curves constitute a continuous structural path based on indicators such as the endpoint positions, tangent directions, and local curvatures. Through interpolation, fitting, or segment connection, these curves can be concatenated into a set of structurally continuous and semantically consistent curves. This set of continuous curves can be viewed as a global representation of risk mutation boundaries obtained from the perspectives of different local models, possessing globally sensitive structural characteristics under multi-model integration.

[0099] According to embodiments of this application, a complete risk bifurcation graph can be constructed based on a set of continuous curves, which can adapt to complex enterprise groups, industry types or data sparse situations, thereby realizing the identification of key variables and the tracking of mutations in a high-dimensional input space.

[0100] In the embodiments of this application, in order to improve the structural continuity and geometric consistency of bifurcation curves generated under multiple local Gaussian process regression models in the input variable space, a pseudo-arc length numerical continuity method can be introduced during the construction of the risk bifurcation diagram to perform guided splicing of the bifurcation curves. Specifically, in the process of splicing multiple bifurcation curves in the input variable space according to numerical continuity, the initial target point position of the next target bifurcation curve can be predicted by linear extrapolation along the tangential direction in the two-dimensional perturbation subspace based on the last target point of the currently constructed bifurcation curve (i.e., the input variable coordinates corresponding to the end point of the curve), using the pseudo-arc length rule.

[0101] The pseudo-arc length method is a solution strategy for numerical path extension problems. Its core idea is to maintain tangential consistency near the end of the current path and estimate the continuation direction of potential paths through small step extrapolation. Based on this principle, the system can use the tangent at the end of the current bifurcation curve as a reference direction, combined with the local coordinate system of the input variable space, to predict the initial position of a potential target point after fine-tuning the values ​​of the first input variables. This position serves as a candidate starting point for connecting the next bifurcation curve, maintaining the overall geometric continuity of the curve while also improving the physical rationality and numerical stability of the splicing.

[0102] After obtaining the initial target point location, its corresponding first input variable can be set to a fixed value and substituted into the corresponding local Gaussian process regression model. The first partial derivative of the default prediction value can then be calculated along the direction of the second input variable. By solving for the points where the partial derivative is zero in this direction, the true prediction target point within the region can be determined. This ensures that the splicing point is not only spatially reasonable but also satisfies the risk mutation characteristics of local extrema or turning points in the model output gradient structure, thereby maintaining the theoretical consistency of the bifurcation curve definition.

[0103] Furthermore, the newly determined predicted target point can be sequentially connected to the endpoint target point of the current bifurcation curve through a spatial connection operation, completing the splicing between the current bifurcation curve and the target bifurcation curve. This connection process can be smoothed using methods such as interpolation, spline fitting, or gradient constraint optimization to ensure the continuity and differentiability of the connection between curves and avoid structural jumps or analysis breakpoints.

[0104] According to the embodiments of this application, not only is the structural splicing of bifurcation curves between multiple local models achieved, but the fine-grained expression of local risk-sensitive points by each local model is also preserved. This mechanism significantly enhances the overall coherence of the graph structure and the fusion consistency of multi-model outputs when constructing the final risk bifurcation graph, providing a modeling foundation for high-dimensional risk structure analysis, strategy sensitivity identification, and visualization interaction. In scenarios dealing with low-dimensional variables but complex model responses (such as bivariate risk analysis and identification of financial sensitivity factors in micro and small enterprises), this method can efficiently splice local curves and construct continuous risk mutation boundaries, effectively improving the overall performance of the system in terms of interpretability, stability, and adaptability.

[0105] By integrating the response results of multiple local models, the risk bifurcation diagram retains the modeling ability of each local model for different subdomains of the data space, avoiding the dilution of risk signals caused by global model averaging. At the same time, it reveals the boundary behavior of input variables under different risk contexts, providing a refined explanation for policy generation and supporting the generation of risk mitigation paths through reverse deduction.

[0106] Preferably, embodiments of this application further include taking a stop operation in response to the prediction uncertainty of the current local Gaussian process regression model exceeding a preset threshold, to enhance the stability and accuracy of the model. Specifically, if during operation, the system detects that the prediction uncertainty of the current local Gaussian process regression model exceeds the preset threshold, the current target point of the current bifurcation curve can be used as the stop point, and the prediction expansion of the local model in that specific input space region will be terminated. The purpose of this operation is to avoid continuing to make predictions that amplify errors in regions with high uncertainty, ensuring the high reliability of the risk assessment model at key decision points.

[0107] Once the stopping point is determined, a target local Gaussian process regression model can be established based on its position in the input variable space. This model serves as the starting point for the next bifurcation curve, used to generate the target bifurcation curve. The newly established model focuses primarily on the characteristics of the current stopping point and its surrounding region, thereby providing more accurate default predictions within the new input variable space and avoiding unnecessary risks caused by over-inference. In this way, the system can dynamically adjust the model's adaptability, intelligently switching to a more suitable local modeling approach when high prediction uncertainty exists, thus optimizing the accuracy and robustness of the entire risk prediction process.

[0108] According to embodiments of this application, by introducing an uncertainty control mechanism, the system can effectively avoid over-expanding prediction results in unclear areas, ensuring the stability of the model and the accuracy of risk identification.

[0109] Figure 4 The illustration shows a schematic diagram of a method for risk prediction of an object to be evaluated according to some exemplary embodiments of this application.

[0110] like Figure 4 As shown, the method for risk prediction of the object to be evaluated may include operations S410 to S420.

[0111] In operation S410, multiple predicted state points are formed in the input variable space based on the multiple default prediction values.

[0112] Specifically, based on the multiple default prediction values ​​obtained in operation S220, they can be mapped to the corresponding input variable space to form a set of predicted state points reflecting the model's response under the current financial characteristics. Each predicted state point corresponds to the input position of a local model and its output default probability, essentially representing a snapshot of the state of the object to be evaluated from multiple local risk perspectives.

[0113] In operation S420, risk prediction of the object to be evaluated is performed based on the judgment of whether the multiple predicted state points fall into the mutation boundary region of the risk bifurcation diagram.

[0114] Furthermore, the predicted state points can be matched with the risk bifurcation diagram to determine whether these state points fall within the abrupt change boundary region indicated by the bifurcation diagram. The abrupt change boundary region is a highly sensitive interval formed by connecting multiple bifurcation curves, reflecting the location where the model output undergoes a nonlinear jump under the perturbation of input variables. If the predicted state point falls into this region, it indicates that the current financial characteristic data of the object under evaluation is in a risk-sensitive segment, and even if the current predicted value may still be in the low to medium risk zone, there is a potential risk that a small change in variables could trigger a sharp increase in the probability of default.

[0115] In embodiments of this application, the system may further introduce a mutation region tolerance threshold mechanism to dynamically determine the strength of a state point's fall into the bifurcation curve based on the shortest distance between the state point and the bifurcation curve or the boundary density of the region it occupies. For example, if a state point is located within the mutation boundary buffer (e.g., less than a preset distance), it is marked as "near a high-sensitivity boundary" and needs to be dynamically monitored in conjunction with other variable trends; if the state point is deeply embedded inside the bifurcation graph, the system can immediately issue a high-risk warning and mark the object as being in a "structurally unstable" input configuration.

[0116] Through the above methods, the system not only considers the magnitude of the default prediction value itself, but also integrates the geographical structural location of the prediction value in the input space with the risk boundary structure of the bifurcation diagram, thus achieving a leap from point value judgment to spatial structure judgment.

[0117] According to embodiments of this application, the accuracy and interpretability of financial risk assessment are improved by introducing multiple local Gaussian process regression models, constructing bifurcation curves, and generating risk bifurcation diagrams. By dynamically stitching together local bifurcation curves, accurately identifying risk mutation boundaries, and combining a prediction uncertainty control mechanism, the system can meticulously identify sensitive regions in the input variable space and intelligently adjust modeling strategies under conditions of high uncertainty, thereby providing more stable and reliable risk prediction results. The method provided by embodiments of this application not only improves adaptability to complex data structures but also optimizes risk warning, strategy formulation, and risk intervention decisions, making it suitable for refined assessment of micro and small enterprises, non-standardized objects, and highly dynamic risk scenarios.

[0118] Corresponding to the above-described risk prediction method, embodiments of this application also provide a risk prediction device.

[0119] Figure 5 A schematic block diagram of a risk prediction device according to an embodiment of this application is shown.

[0120] like Figure 5 As shown, the risk prediction device 500 of this embodiment includes a data acquisition module 510, a default prediction value acquisition module 520, a risk bifurcation diagram acquisition module 530, and a risk prediction module 540.

[0121] The data acquisition module 510 can be used to acquire financial characteristic data of the object to be evaluated. In one embodiment, the data acquisition module 510 can be used to perform the operation S210 described above, which will not be repeated here.

[0122] The default prediction value acquisition module 520 can be used to input the financial characteristic data into multiple pre-built local Gaussian process regression models and output multiple default prediction values. In one embodiment, the default prediction value acquisition module 520 can be used to perform the operation S220 described above, which will not be repeated here.

[0123] The risk bifurcation diagram acquisition module 530 can be used to obtain a risk bifurcation diagram based on the multiple local Gaussian process regression models. The risk bifurcation diagram is used to identify the input variable threshold that causes a sudden change in the default prediction value under the conditions of disturbance in the financial characteristic data. In one embodiment, the risk bifurcation diagram acquisition module 530 can be used to perform the operation S230 described above, which will not be repeated here.

[0124] The risk prediction module 540 can be used to predict the risk of the object to be evaluated based on the multiple default prediction values ​​and the risk bifurcation diagram. In one embodiment, the risk prediction module 540 can be used to perform the operation S240 described above, which will not be repeated here.

[0125] According to an embodiment of this application, the risk bifurcation diagram acquisition module 530 can also be used to construct an input variable space based on multiple input variables in the financial feature data; generate a corresponding bifurcation curve for each local Gaussian process regression model and the input variable space; and construct the risk bifurcation diagram based on multiple bifurcation curves corresponding to the multiple local Gaussian process regression models.

[0126] According to an embodiment of this application, the risk bifurcation diagram acquisition module 530 can also be used to determine the target point where the partial derivative of the default prediction value of the local Gaussian process regression model with respect to the direction of the second input variable is zero under the condition that the first input variable is set to different fixed values; and to connect the target point continuously in the input variable space to generate the corresponding bifurcation curve.

[0127] According to embodiments of this application, the risk bifurcation diagram acquisition module 530 can also be used to splice the multiple bifurcation curves in the input variable space according to numerical continuity to form a continuous curve set; and

[0128] Based on the set of continuous curves, the risk bifurcation diagram is constructed.

[0129] According to an embodiment of this application, the risk bifurcation diagram acquisition module 530 can also be used to predict the initial target point position of the target bifurcation curve based on the final target point of the current bifurcation curve by performing linear extrapolation in the input variable space using a pseudo-arc length numerical continuity method; under the condition that the first input variable of the initial target point position is set to a fixed value, determine the prediction target point where the partial derivative of the default prediction value of the corresponding local Gaussian process regression model with respect to the direction of the second input variable is zero; and connect the prediction target point and the target point of the current bifurcation curve in sequence to splice the current bifurcation curve and the target bifurcation curve.

[0130] According to an embodiment of this application, the risk bifurcation diagram acquisition module 530 can also be used to, in response to the prediction uncertainty of the current local Gaussian process regression model exceeding a preset threshold, take the current target point of the current bifurcation curve as a stopping point; and establish a target local Gaussian process regression model based on the position of the stopping point in the input variable space, wherein the target local Gaussian process regression model is used to generate the target bifurcation curve.

[0131] According to an embodiment of this application, the risk prediction module 540 can also be used to respond to the fact that the prediction uncertainty of the current local Gaussian process regression model exceeds a preset threshold, and to take the current target point of the current bifurcation curve as a stop point; based on the position of the stop point in the input variable space, to establish a target local Gaussian process regression model, which is used to generate the target bifurcation curve.

[0132] According to embodiments of this application, any multiple modules among the data acquisition module 510, default prediction value acquisition module 520, risk bifurcation map acquisition module 530, and risk prediction module 540 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the data acquisition module 510, default prediction value acquisition module 520, risk bifurcation map acquisition module 530, and risk prediction module 540 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the data acquisition module 510, the default prediction value acquisition module 520, the risk bifurcation diagram acquisition module 530, and the risk prediction module 540 may be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0133] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a risk prediction method according to an embodiment of this application.

[0134] like Figure 6 As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0135] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0136] According to embodiments of this application, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0137] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0138] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0139] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the risk prediction method provided in the embodiments of this application.

[0140] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0141] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0142] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0143] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0144] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, 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 indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated 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 or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0145] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0146] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A risk prediction method, characterized in that, The method includes: Obtain financial characteristic data of the object to be evaluated; The financial characteristic data are input into multiple pre-built local Gaussian process regression models, and multiple default prediction values ​​are output. A risk bifurcation diagram is obtained based on the multiple local Gaussian process regression models. This risk bifurcation diagram is used to identify the threshold of input variables that cause abrupt changes in the default prediction value under the conditions of disturbances in the financial characteristic data. Based on the multiple default prediction values ​​and the risk bifurcation diagram, the risk prediction of the object to be evaluated is performed.

2. The method according to claim 1, characterized in that, The risk bifurcation diagram obtained based on the multiple local Gaussian process regression models includes: Based on multiple input variables in the aforementioned financial feature data, an input variable space is constructed; For each local Gaussian process regression model and the input variable space, generate the corresponding bifurcation curve; and The risk bifurcation diagram is constructed based on the multiple bifurcation curves corresponding to the multiple local Gaussian process regression models.

3. The method according to claim 2, characterized in that, The plurality of input variables includes at least a first input variable and a second input variable. For each local Gaussian process regression model and the input variable space, generating a corresponding bifurcation curve includes: Under the condition that the first input variable is set to different fixed values, determine the target point where the partial derivative of the local Gaussian process regression model with respect to the default prediction value in the direction of the second input variable is zero; and The target points are continuously connected in the input variable space to generate the corresponding bifurcation curve.

4. The method according to claim 2 or 3, characterized in that, The risk bifurcation diagram is constructed based on the multiple bifurcation curves corresponding to the multiple local Gaussian process regression models, including: The multiple bifurcated curves are spliced ​​together in the input variable space according to numerical continuity to form a set of continuous curves; and Based on the set of continuous curves, the risk bifurcation diagram is constructed.

5. The method according to claim 4, characterized in that, The step of splicing the multiple bifurcated curves in the input variable space according to numerical continuity includes: Based on the final target point of the current bifurcation curve, the initial target point position of the target bifurcation curve is predicted by linear extrapolation in the input variable space using the pseudo-arc length numerical continuity method. Under the condition that the first input variable of the initial target point location is set to a fixed value, the prediction target point where the partial derivative of the corresponding local Gaussian process regression model with respect to the default prediction value in the direction of the second input variable is zero is determined; and The predicted target point and the target point of the current bifurcation curve are connected sequentially to splice the current bifurcation curve and the target bifurcation curve.

6. The method according to claim 2, characterized in that, The risk prediction of the object to be evaluated based on the multiple default prediction values ​​and the risk bifurcation diagram includes: Based on the multiple default prediction values, multiple prediction state points are formed in the input variable space; and Risk prediction of the object to be evaluated is performed based on the determination of whether the multiple predicted state points fall into the mutation boundary region of the risk bifurcation diagram.

7. The method according to claim 5, characterized in that, The method further includes: In response to the prediction uncertainty of the current local Gaussian process regression model exceeding a preset threshold, the current target point of the current bifurcation curve is taken as the stopping point; Based on the position of the stopping point in the input variable space, a target local Gaussian process regression model is established, which is used to generate the target bifurcation curve.

8. A risk prediction device, characterized in that, The device includes: The data acquisition module is used to: acquire financial characteristic data of the object to be evaluated; The default prediction value acquisition module is used to: input the financial characteristic data into multiple pre-built local Gaussian process regression models respectively, and output multiple default prediction values; The risk bifurcation diagram acquisition module is used to: obtain a risk bifurcation diagram based on the multiple local Gaussian process regression models, wherein the risk bifurcation diagram is used to identify the input variable threshold that causes abrupt changes in the default prediction value under the conditions of disturbance in the financial feature data; and The risk prediction module is used to: predict the risk of the object to be evaluated based on the multiple default prediction values ​​and the risk bifurcation diagram.

9. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

Images