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

Abstract

The invention provides a risk prediction method and device, equipment, a storage medium and a program product, and relates to the field of artificial intelligence. The method comprises the steps of obtaining to-be-processed information, wherein the to-be-processed information comprises multiple types of feature information of a target object; processing the to-be-processed information based on the plurality of risk prediction models to obtain a prediction risk degree output by each risk prediction model; current weight distribution information and a risk threshold value are read from a preset storage space, and the weight distribution information is used for representing the weight corresponding to each risk prediction model; performing weighted fusion on the prediction risk degrees corresponding to the risk prediction models according to the weight distribution information to obtain a fused prediction risk degree; and determining a risk prediction result corresponding to the target object according to the fused prediction risk degree and the risk threshold. According to the method, the technical effect of improving the risk prediction accuracy is achieved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a risk prediction method, device, equipment, storage medium and program product. Background Art

[0002] With the development of small and micro enterprises, banks have launched a new generation of quick business loan products to solve their financing problems. However, due to the limited scale, large financial fluctuations and sensitivity to market changes of small and micro enterprises, banks face huge challenges in avoiding fraud in all-online operations and achieving credit risk control under a fully open customer acquisition model.

[0003] In the existing credit risk identification and prediction, banks and other financial institutions generally use risk prediction models to predict the risks of credit products. The risk prediction model generally includes a single algorithm, such as logistic regression model, decision tree model, random forest and deep learning model.

[0004] However, existing credit risk identification and prediction models have certain limitations. Different financial market risk prediction models can only target specific market types and cannot be applied to all types of markets, which leads to low accuracy of model predictions. Summary of the invention

[0005] The present application provides a risk prediction method, device, equipment, storage medium and product to solve the technical problem of poor risk prediction accuracy.

[0006] In a first aspect, the present application provides a risk prediction method, comprising:

[0007] Acquire information to be processed, where the information to be processed includes multiple types of feature information of the target object;

[0008] Based on multiple risk prediction models, the information to be processed is processed respectively to obtain the predicted risk degree output by each risk prediction model;

[0009] Read the current weight distribution information and risk threshold from the preset storage space, wherein the weight distribution information is used to characterize the weights corresponding to each risk prediction model;

[0010] According to the weight distribution information, the predicted risk levels corresponding to each risk prediction model are weighted and integrated to obtain the integrated predicted risk level;

[0011] According to the fused predicted risk level and risk threshold, the risk prediction result corresponding to the target object is determined.

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

[0013] An acquisition module is used to acquire information to be processed, where the information to be processed includes various types of feature information of the target object;

[0014] A first processing module is used to process the information to be processed based on multiple risk prediction models respectively to obtain the predicted risk level output by each risk prediction model;

[0015] A reading module is used to read the current weight distribution information and risk threshold from a preset storage space, wherein the weight distribution information is used to characterize the weights corresponding to each risk prediction model;

[0016] A fusion module is used to perform weighted fusion on the predicted risk levels corresponding to the various risk prediction models according to the weight distribution information to obtain the fused predicted risk level;

[0017] The second processing module is used to determine the risk prediction result corresponding to the target object according to the fused predicted risk level and risk threshold.

[0018] In a third aspect, the present application provides a risk prediction device, including: a memory, a processor;

[0019] Memory stores computer-executable instructions;

[0020] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the implementation of the first aspect as described above.

[0021] In a fourth aspect, the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the implementation method of the first aspect as described above.

[0022] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, is used to implement the implementation of the first aspect as above.

[0023] The present application provides a risk prediction method, device, equipment, storage medium and program product, which integrate various types of feature information as information to be processed to avoid one-sided evaluation caused by a single factor. Multiple risk prediction models are used to analyze the information to be processed from different angles, and then the weights corresponding to each risk prediction model are used to weightedly fuse the predicted risk levels output by each risk prediction model, thereby improving the accuracy and reliability of the prediction. And by flexibly adjusting the weight distribution information and risk threshold according to actual conditions, it can better adapt to different scenarios and needs. Therefore, the above method can not only meet the actual use needs of different scenarios, but also improve the technical effect of the accuracy of risk prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0025] Figure 1 A schematic diagram of a risk prediction method provided in an embodiment of the present application Figure 1 ;

[0026] Figure 2 A schematic diagram of a risk prediction method provided in an embodiment of the present application Figure 2 ;

[0027] Figure 3 A schematic diagram of a risk prediction method provided in an embodiment of the present application Figure 3 ;

[0028] Figure 4 A schematic diagram of a risk prediction method provided in an embodiment of the present application Figure 4 ;

[0029] Figure 5 A schematic diagram of a risk prediction method provided in an embodiment of the present application Figure 5 ;

[0030] Figure 6 A schematic diagram of a risk prediction method provided in an embodiment of the present application Figure 6 ;

[0031] Figure 7 A schematic diagram of the structure of a risk prediction device provided in an embodiment of the present application;

[0032] Figure 8 A schematic diagram of the structure of a risk prediction device provided in an embodiment of the present application.

[0033] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0034] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0035] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0036] In addition, this application involves conducting big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.), and using artificial intelligence technology to make automated decisions, and making technical solutions that have a significant impact on personal rights and interests based on the results of automated decisions. The application provides users with corresponding operation entrances for them to choose to agree or reject the results of automated decisions; if the user chooses to reject, the expert decision-making process will be entered.

[0037] It should be noted that the risk prediction method, device, equipment, storage medium and program product provided in the present application can be used in the field of artificial intelligence, and can also be used in any field other than artificial intelligence. The application field of the risk prediction method, device, equipment, storage medium and program product in the present application is not limited.

[0038] In recent years, my country has paid great attention to and supported the growth of small and micro enterprises. Through a series of measures, it has reduced their operating costs, enhanced their market competitiveness, and created a favorable development environment for them. In order to further support the development of small and micro enterprises and solve their financing problems, banks and other financial institutions have launched a new generation of quick loan products. This loan product can effectively meet the changes in the capital needs of small and medium-sized enterprises during their development. Therefore, it has become a key task to predict the risks of small and micro enterprises to quick loan products.

[0039] In the existing credit risk identification and prediction process, banks and other financial institutions usually rely on risk prediction models to assess the risk of credit products. These risk prediction models mainly use single algorithms, such as logistic regression models, decision tree models, random forests, and deep learning models. The logistic regression model has strong interpretability but limited ability to handle nonlinear relationships; the decision tree model is intuitive and easy to understand but prone to overfitting when dealing with high-dimensional data; the random forest model is highly robust but has difficulty dealing with highly complex data structures; the deep learning model is excellent in processing complex data but requires a large amount of data and computing resources and has poor interpretability.

[0040] Therefore, since different financial market risk prediction models can only target specific market types and cannot be applied to all types of markets, the accuracy of credit risk identification and prediction models is low.

[0041] The present application provides a risk prediction method, device, equipment, storage medium and program product, which integrates multiple types of feature information as information to be processed, uses multiple risk prediction models to analyze the information to be processed from different angles, obtains the predicted risk level output by each risk prediction model, and performs weighted fusion based on weight distribution information to obtain the fused predicted risk level. Then, according to the fused predicted risk level and risk threshold, the risk prediction result corresponding to the target object is determined, aiming to solve the above technical problems of the prior art.

[0042] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0043] Figure 1 A schematic diagram of a risk prediction method provided in an embodiment of the present application Figure 1 ,like Figure 1 As shown, the method includes:

[0044] S101. Obtain information to be processed.

[0045] In this embodiment, the information to be processed includes multiple types of characteristic information of the target object. The multiple types of characteristic information include, for example, basic information, financial information, preference information, asset information, and behavior information of the target object. Among them, the basic information of the target object includes, for example, the target object number, account name, age, years of establishment of the enterprise, marital status, family members and family members' assets, liabilities, credit, income information, residential / business address, contact information, personal occupation / enterprise industry and current risk value of the industry, education, ID number, etc. Financial information includes, for example, operating income and its growth rate, main sources of income, tax records, acquisition data, financial statements, liabilities, etc. Preference information includes, for example, credit card overdraft preferences, consumption preferences, asset allocation preferences, etc. These preferences reflect the consumption habits, risk tolerance and asset management concepts of the target object, and are of reference value for assessing fraud and default risks. Asset information includes, for example, real estate information, bank deposits, other financial assets, vehicle information, etc. These asset information are used to analyze the asset status and debt repayment ability of the target object. Behavioral information includes, for example, transaction behavior, consumption behavior, and arrears behavior. These behavioral information directly reflect the target object's capital flow, consumption habits, repayment willingness, and potential fraud risks.

[0046] It should be noted that the target object number can be used for data integration and tracking; the account name, age and years of establishment of the company provide a basic portrait of the target object, which is of fundamental significance for assessing the stability and credit risk of the target object; marital status affects the financial stability and debt-bearing capacity of the family, and is an important factor in assessing the risk of default; the borrower's family members and their assets, liabilities, credit and income information help to fully understand the family's financial situation, and then help to assess the overall debt repayment ability; residential / business address and contact information are used to verify the authenticity of the target object information and follow up; the individual's occupation / enterprise industry and the current risk value of the industry can directly affect the target object's business stability and debt repayment ability; academic qualifications indirectly reflect the target object's knowledge level and potential repayment willingness; ID number is used for identity verification and cross-verification with public databases; operating income and its growth rate, and main sources of income directly reflect the customer's business status and profitability; tax records are used to assess the customer's tax compliance and business status; collection data is an important indicator for evaluating its business activity and stability; financial statements are used to provide detailed financial status analysis, which is an important basis for evaluating the customer's debt repayment ability; debt situation is used to comprehensively evaluate the customer's debt burden and debt repayment ability.

[0047] In one possible implementation, the information to be processed needs to be preprocessed and feature converted. The preprocessing process includes cleaning and integration of the information to be processed, for example, missing value processing: for missing values ​​in key data such as financial statements, select appropriate filling strategies based on business rules and data distribution characteristics. For data that is seriously missing and cannot be reasonably filled, consider treating it as a separate feature category or eliminating it. Outlier detection and processing: use statistical methods (such as box plots, Z-score, etc.) or business rule-based methods to identify outliers. For reasonable outliers (such as occasional large transactions), their original values ​​can be retained; for unreasonable outliers (such as data entry errors), they are corrected or eliminated. Integration and deduplication of information to be processed: integrate the information to be processed from different data sources to ensure the integrity and consistency of the information to be processed. At the same time, remove duplicate records to reduce the impact of redundant information to be processed on the training of each risk prediction model. The feature conversion process includes transformation and standardization of information to be processed: for example, categorical variable encoding: for categorical variables such as marital status, one-hot encoding or label encoding is used for conversion. For ordered categorical variables (such as educational level), consider using ordered coding to retain their order information. Standardization / normalization of numerical variables: For numerical variables such as financial ratios and transaction amounts, select appropriate standardization or normalization methods based on their distribution characteristics. This helps to eliminate the impact of different dimensions on the training of each risk prediction model and accelerate the convergence speed of each risk prediction model. Time series data processing: For time series data such as transaction behavior and consumption behavior, time series analysis and processing are performed to extract useful feature information. For example, calculate the statistics of the time series (mean, standard deviation, extreme value, etc.), perform trend analysis or seasonal adjustment, etc.

[0048] S102: Process the information to be processed respectively based on multiple risk prediction models to obtain the predicted risk level output by each risk prediction model.

[0049] In this embodiment, the risk prediction of the market is usually uncertain, and the target prediction effect cannot be achieved by only using a single model. Therefore, in the process of improving the accuracy of risk prediction, multiple risk prediction models can be used. Each model processes the input information to be processed according to its specific algorithm and rules, and outputs the predicted risk level of the target object. The predicted risk level can be, for example, a value or a level to indicate the possibility of the target object facing risk.

[0050] Optionally, for example, multiple risk prediction models matching the information to be processed can be selected from a preset model library based on the type of feature information in the information to be processed and the data size of the information to be processed. For example, there can be multiple models in the model library to adapt to different information to be processed. For example, if the information to be processed includes time series data, a long short-term memory model can be used, so as to automatically adapt the model according to the current information type to further improve the effect.

[0051] In one possible implementation, multiple risk prediction models include at least two of the Long Short-Term Memory (LSTM) model, the Random Forest algorithm, and the eXtreme Gradient Boost (XGBoost) algorithm. Among them, the Long Short-Term Memory model introduces three key components: the forget gate, the input gate, and the output gate. These gates control the inflow, retention, and outflow of information to be processed through mathematical formulas. The forget gate is used to determine which information to be processed is discarded from the cell state, and the new information to be processed is obtained through the following formula:

[0052]

[0053] in, is the output of the forget gate, σ is Activation function, is the weight matrix, is the bias term, is the hidden state at the previous time step, is the input for the current time step.

[0054] Input gate and candidate cell state: jointly decide which new information to be processed will be added to the cell state, which is achieved through the following formula:

[0055]

[0056]

[0057] in, is the output of the input gate, is a candidate cell state, and is the weight matrix, and is the bias term.

[0058] Cell state update: Combine the results of the forget gate and the input gate to update the cell state, which is achieved through the following formula:

[0059]

[0060] in, is the updated cell state, Represents element-wise multiplication.

[0061] Output gate: controls which information to be processed will be output from the cell state to the hidden state, which is implemented by the following formula:

[0062]

[0063]

[0064] in, is the output of the output gate, is the hidden state at the current time step.

[0065] This mechanism enables the LSTM model to effectively learn and memorize long-term dependencies. The cell state of the LSTM model acts as a long-term memory unit to transmit information along the time series. Through the synergy of the forget gate and the input gate, historical information useful for the prediction task is retained while irrelevant noise is discarded, thereby improving the accuracy of the prediction.

[0066] The data on the customer's business conditions, transaction behaviors, etc. in the information to be processed have obvious time series characteristics. The LSTM model can use these time series data to learn and capture the time dependency and change trends in the data, and then predict future risk conditions.

[0067] The random forest algorithm reduces overfitting and improves prediction stability by building multiple decision trees; XGBoost provides an optimized gradient boosting algorithm to efficiently process large-scale data.

[0068] S103: Read current weight distribution information and risk threshold from a preset storage space.

[0069] In this embodiment, the weight distribution information is used to characterize the weights corresponding to each risk prediction model, clarify the proportion of each risk prediction model in the overall model, and reflect the current importance of different risk prediction models. The risk threshold can be used as a standard value to determine whether the risk exceeds a specific limit, and is used to characterize the tolerance of banks and other financial institutions or enterprises to risk.

[0070] S104. Perform weighted fusion on the predicted risk levels corresponding to the risk prediction models according to the weight distribution information to obtain a fused predicted risk level.

[0071] In this embodiment, the predicted risk levels output by each risk prediction model are weighted calculated based on the weight distribution information, that is, the predicted risk levels of each model are adjusted according to the corresponding weights and then merged to obtain a comprehensive, integrated predicted risk level. This integrated predicted risk level more comprehensively reflects the overall risk status of the target object.

[0072] It should be noted that the weighted fusion method refers to assigning a weight value to the predicted risk level corresponding to each risk prediction model, which determines the importance of each predicted risk level when calculating the average value. By multiplying each predicted risk level with its corresponding weight value and then summing the results, the fused predicted risk level is obtained.

[0073] In one possible implementation, taking multiple risk prediction models including the Long Short-Term Memory (LSTM) model, the random forest algorithm and the eXtreme Gradient Boost (XGBoost) algorithm as examples, if the predicted risk level output by the LSTM model is a, the predicted risk level output by the random forest algorithm is b, and the predicted risk level output by the XGBoost algorithm is c, and the weight distribution of the LSTM model, the random forest algorithm and the XGBoost algorithm is α:β:γ, then the predicted risk levels corresponding to the risk prediction models are weighted and fused according to the weight distribution information, and the fused predicted risk level is .

[0074] S105. Determine the risk prediction result corresponding to the target object according to the fused predicted risk level and risk threshold.

[0075] In this embodiment, the predicted risk level after fusion is compared with the risk threshold. If the predicted risk level after fusion exceeds the risk threshold, it is determined that the target object has a high risk and a corresponding risk prediction result is obtained; if it does not exceed, it is determined that the risk is low or within an acceptable range and a corresponding risk prediction result is obtained.

[0076] Optionally, taking a benchmark risk threshold as an example, low risk: predicted risk level ≤ 0.1; medium risk: 0.1 < predicted risk level ≤ 0.5; high risk: predicted risk level > 0.5. When the predicted risk level ≤ 0.1, the corresponding risk prediction result is low risk; when 0.1 < predicted risk level ≤ 0.5, the corresponding risk prediction result is medium risk; when the predicted risk level > 0.5, the corresponding risk prediction result is high risk.

[0077] In one possible implementation, after the risk prediction results are obtained, the risk prediction results can be pushed to business personnel via mobile phone text messages or at the bank front counter, so that the business personnel can be notified in time to pay attention to the risks of the target objects.

[0078] In one possible implementation, the method in the above embodiment can be applied to other scenarios of credit scenarios, and in addition to being applied to the financial field, it can also be applied to other fields, such as predicting natural disaster risks, predicting abnormal risks of data processing systems, etc. The corresponding input data can also be of multiple types, such as meteorological data, historical disaster data, and performance indicator data. The final results are natural disaster risk prediction and abnormal detection result detection.

[0079] The risk prediction method provided in the embodiment of the present application can more comprehensively understand the target object and avoid one-sided estimation caused by a single factor by integrating various types of feature information as information to be processed. Based on multiple risk prediction models, the information to be processed is processed separately to obtain the predicted risk degree output by each risk prediction model. By using multiple risk prediction models to analyze from different angles, and then through weighted fusion, the accuracy and reliability of the prediction are improved. And by flexibly adjusting the weight distribution information and risk threshold according to the actual situation, it is better to adapt to different scenarios and needs. Therefore, the above method can not only meet the actual use needs of different scenarios, but also improve the accuracy of risk prediction.

[0080] Figure 2 A schematic diagram of a risk prediction method provided in an embodiment of the present application Figure 2 ,like Figure 2 As shown, in this embodiment Figure 1 Based on the embodiment, a possible implementation method for determining the weight distribution information is described in detail, and the method includes:

[0081] S201. Obtain verification information corresponding to a plurality of historical objects within a preset period at every preset period.

[0082] In this embodiment, a time period is defined as a preset period, such as monthly, quarterly, etc., and verification information corresponding to multiple historical objects within the preset period is collected, wherein the verification information corresponding to the historical object includes a label corresponding to the historical object and multiple types of characteristic information, and the label is used to indicate whether a risk event has occurred. The multiple types of characteristic information are historical basic information, historical financial information, historical preference information, historical asset information, and historical behavior information corresponding to multiple historical objects within the preset period.

[0083] S202. Verify each risk prediction model according to the verification information corresponding to multiple historical objects within a preset period to obtain the prediction accuracy corresponding to each risk prediction model.

[0084] In this embodiment, the actual risk level reflected by the label of each historical object within a preset period is compared with the prediction results output by each risk prediction model, and the prediction accuracy of each risk prediction model within this preset period is calculated based on the comparison results.

[0085] S203: Determine an auxiliary evaluation index to be calculated from a plurality of preset auxiliary evaluation indexes according to the type of feature information in the verification information.

[0086] In this embodiment, according to the type of characteristic information in the verification information, an indicator suitable for the current situation is selected from a plurality of pre-set auxiliary evaluation indicators as the auxiliary evaluation indicator to be calculated. The plurality of auxiliary evaluation indicators include the area under the receiver operating characteristic curve ROC and the discrimination evaluation index. The receiver operating characteristic curve ROC is used to measure the classification ability of the risk prediction model, and the discrimination evaluation index is used to evaluate the ability of the risk prediction model to distinguish different risk levels.

[0087] In one possible implementation, different prediction scenarios use different types of feature information and different performance indicators are expected to be achieved. If the type of feature information is a categorical feature, the categorical feature includes, for example, discrete values ​​such as object category and transaction type. For models that rely on categorical features, the receiver operating characteristic curve (ROC) or discrimination evaluation index is generally used as a common auxiliary evaluation index because they can effectively measure the classification ability of the model.

[0088] In a possible implementation, if the type of feature information is a numerical feature, a time series feature, etc. Numerical features are continuous values ​​such as transaction amounts and account balances. Time series features are transaction records, repayment records, etc. If the model mainly relies on numerical features, the mean square error (MSE) can be selected as an auxiliary evaluation indicator. If the feature information involves time series data, it is necessary to consider time-related evaluation indicators, such as the results of time series cross-validation as auxiliary evaluation indicators.

[0089] S204. Calculate auxiliary evaluation indicators corresponding to each risk prediction model based on verification information corresponding to multiple historical objects within a preset period.

[0090] In this embodiment, the verification information corresponding to multiple historical objects within a preset period is input into each risk prediction model to obtain the corresponding predicted risk level, and the predicted risk level is compared with the actual risk level reflected by the label. Based on the comparison results, the corresponding auxiliary evaluation indicators of each risk prediction model are calculated.

[0091] In a possible implementation, the multiple auxiliary evaluation indicators include the area under the receiver operating characteristic curve (ROC) and the discrimination evaluation indicator. Taking the area under the receiver operating characteristic curve (ROC) as an example, the ROC curve can be calculated based on the comparison results, and the area under the ROC can be calculated. The higher the area, the better the classification performance of the corresponding risk prediction model.

[0092] S205. Update the weights corresponding to the risk prediction models according to the prediction accuracy and auxiliary evaluation indicators corresponding to the risk prediction models.

[0093] In this embodiment, the weights of the prediction accuracy and auxiliary evaluation indicators corresponding to each risk prediction model can be set according to the scenario requirements or experience, and the prediction accuracy and auxiliary evaluation indicators are weighted and summed according to the corresponding weights to obtain the corresponding comprehensive score. According to the comprehensive score, the weight of each risk prediction model in the combined model is adjusted. The risk prediction model with a high score obtains a higher weight, and the risk prediction model with a low score has a lower weight.

[0094] A risk prediction method provided by an embodiment of the present application can dynamically adapt to changes in data distribution and risk environment by regularly obtaining verification information and updating model weights. This continuous optimization ensures that each risk prediction model remains efficient and accurate in different time periods. By using multiple evaluation indicators, the model can show stronger robustness under different feature types and risk scenarios. By dynamically adjusting the weights of each risk prediction model in the combined model according to its performance, the overall performance of the model combination can be optimized, ensuring that the combined model can fully utilize the advantages of each risk prediction model and improve the accuracy of risk prediction.

[0095] Figure 3 A schematic diagram of a risk prediction method provided in an embodiment of the present application Figure 3 ,like Figure 3 As shown, in this embodiment Figure 1 Based on the embodiment, a possible implementation method for determining the risk threshold is described in detail, and the method includes:

[0096] S301. Determine risk events occurring within a preset period at every preset period.

[0097] In this embodiment, a time period is defined as a preset period, such as every month, every quarter, etc. At the end of each preset period, it is possible to identify which events within the preset period are risk events through an anomaly detection algorithm, record the identified risk events, and classify them by type or severity.

[0098] S302: Update the risk threshold according to the frequency of occurrence of risk events within a preset period.

[0099] Specifically, if the frequency of risk events in the preset period is lower than the frequency of risk events in the previous period, and the difference is greater than the preset frequency threshold, the risk threshold is lowered, and the degree of reduction is positively correlated with the difference, which means that the greater the difference, the more the risk threshold is lowered. If the frequency of risk events in the previous period is lower than the frequency of risk events in the preset period, and the difference is greater than the preset frequency threshold, the risk threshold is raised, and the degree of increase is positively correlated with the difference, which means that the greater the difference, the more the risk threshold is raised.

[0100] In this embodiment, the occurrence data of risk events within a preset period is obtained. This includes the number of risk events in the previous period and the number of risk events in the current period. The number of risk events in the current period and the number of risk events in the previous period are calculated respectively to obtain the probability of occurrence of risk events in the two periods. The probability of occurrence can be calculated by dividing the number of risk events by the length of the period (for example, the number of days, weeks or months). The frequency of occurrence of risk events in the two periods is compared to determine whether the frequency increases or decreases. The difference in the frequency of occurrence of risk events in the two periods is calculated to determine whether the difference exceeds the preset frequency threshold. If so, the risk threshold is adjusted. The setting of the preset frequency threshold can avoid frequent adjustment of the risk threshold in the current preset period, and the above method can accurately and dynamically adjust the risk threshold according to the changes in the frequency of risk events, so as to better adapt to the risk conditions at different stages.

[0101] S303. Calculate performance indicators corresponding to multiple risk prediction models based on the determined risk events.

[0102] In this embodiment, the performance index of each risk prediction model is calculated based on the actual occurrence of risk events and the prediction results of multiple risk prediction models, which can accurately measure the actual effect and accuracy of each risk prediction model. The performance index can include at least one of the prediction accuracy and auxiliary evaluation index.

[0103] S304. According to the weights corresponding to the risk prediction models, weighted sum is performed on the performance indicators corresponding to the multiple risk prediction models to obtain a comprehensive performance indicator.

[0104] In this embodiment, the weight corresponding to each risk prediction model is obtained, for example, according to the prediction accuracy and auxiliary evaluation index corresponding to each risk prediction model. The performance indicators of multiple risk prediction models are weighted to calculate so as to comprehensively consider the importance and performance of multiple models, and in this way, a comprehensive performance value, i.e., a comprehensive performance index, is obtained to more accurately reflect the overall performance.

[0105] S305. Update the risk threshold according to the comprehensive performance indicator.

[0106] In one possible implementation, if the comprehensive performance indicator corresponding to the preset period is lower than the comprehensive performance indicator of the previous period, and the difference is greater than the preset performance threshold, the risk threshold is lowered, and the degree of reduction is positively correlated with the difference; if the comprehensive performance indicator corresponding to the previous period is lower than the performance indicator of the preset period, and the difference is greater than the preset performance threshold, the risk threshold is lowered, and the degree of reduction is positively correlated with the difference.

[0107] Specifically, the comprehensive performance indicator data within the preset period is obtained. This includes the comprehensive performance indicator within the previous period and the comprehensive performance indicator within the current preset period. The difference between the comprehensive performance indicators within the two periods is calculated. If the comprehensive performance indicator of the current preset period is lower than the comprehensive performance indicator of the previous period, and the difference is greater than the preset performance threshold, the step of reducing the risk threshold is performed, and the risk threshold is reduced in a positive correlation according to the size of the difference. This means that the greater the difference, the more the risk threshold is reduced. If the comprehensive performance indicator in the previous period is lower than the comprehensive performance indicator of the current preset period, and the difference is greater than the preset performance threshold, the step of increasing the risk threshold is performed, and the risk threshold is increased in a positive correlation according to the size of the difference. This means that the greater the difference, the more the risk threshold is increased. By dynamically adjusting the risk threshold using the changes in the comprehensive performance indicators, the organization can respond to the changing risk environment more flexibly. This dynamic adjustment mechanism enables the risk management strategy to be consistent with the actual business environment and risk conditions, thereby improving the effectiveness and adaptability of risk management.

[0108] In a possible implementation, if the comprehensive performance indicator is greater than the comprehensive performance indicator threshold, the risk threshold is updated.

[0109] Specifically, a comprehensive performance index threshold is set, for example, the initial value is set to 0.5, and the comprehensive performance index is compared with the comprehensive performance index threshold. If the comprehensive performance index is greater than the comprehensive performance index threshold, it means that the performance of the risk prediction model is good, and the risk threshold is appropriately increased to reduce the risk of false positives and false negatives. Otherwise, the risk threshold is lowered to increase the sensitivity to potential risks. By setting a comprehensive performance index threshold and dynamically adjusting the comprehensive performance index threshold, it is helpful to discover potential risk problems and improve the accuracy of risk prediction results.

[0110] In a possible implementation, the time length corresponding to the preset period is updated according to the comprehensive performance index and the frequency of occurrence of risk events, wherein the comprehensive performance index is positively correlated with the time length, and the frequency of occurrence is negatively correlated with the time length.

[0111] In this embodiment, if the comprehensive performance index is improved and the frequency of risk events is reduced, the length of the preset period can be increased, which indicates that the risk prediction result is more stable and reliable, and the prediction efficiency is improved. On the contrary, if the comprehensive performance index is reduced or the frequency of risk events is increased, the length of the preset period needs to be shortened for accurate monitoring and adjustment. This method can better adapt to the dynamic changes of risk status by flexibly adjusting the length of the preset period according to actual conditions.

[0112] Optionally, a possible implementation of determining the risk threshold is described: in response to a user changing risk preference information through an interactive interface, the risk threshold is updated according to the risk preference information, wherein the preference information is used to indicate the user's tolerance for risk.

[0113] In this embodiment, interface elements such as sliders, input boxes or selection boxes are designed as interactive interfaces for users to adjust their risk preferences, and risk preference information input or adjusted by users on the interface is collected in real time. The user input is verified to ensure that the input is within a reasonable range, and the user's risk preference information is mapped to specific risk threshold adjustment parameters to update the risk threshold. For example, the user's risk tolerance is converted into a specific numerical change.

[0114] Optionally, the current risk threshold is adjusted based on changes in the market environment. When the market environment changes significantly (such as a deterioration in the economic situation, etc.), the user can change the risk preference information through the interactive interface, and then adjust the risk threshold to adapt to the new environment. The adjustment range can be determined based on the degree of change in the market environment and the business impact. For example, if the deterioration of the economic situation leads to an increase in the customer default rate, the threshold can be significantly lowered to increase risk sensitivity.

[0115] Optionally, a possible implementation method for determining the risk threshold is described: the risk threshold is updated according to the combined effect of risk events and risk preference information.

[0116] Optionally, a possible implementation method for determining the risk threshold is described: the risk threshold is updated according to the frequency of occurrence of risk events within a preset period and performance indicators corresponding to multiple risk prediction models.

[0117] The risk prediction method provided in the embodiment of the present application determines risk events and updates the risk threshold at preset intervals, so that the risk threshold can dynamically adapt to changes in actual risk conditions and better fit the actual situation; the risk threshold is reasonably adjusted based on the comparison of the frequency of risk events with the previous period, and the degree of adjustment is related to the difference, thereby improving the accuracy and rationality of the adjustment; the performance indicators of multiple risk prediction models are calculated and weighted summed to obtain a comprehensive performance indicator, which helps to understand the performance of each risk prediction model, so that the risk threshold can be further optimized according to the comprehensive performance indicator, thereby improving the accuracy and reliability of the overall risk prediction.

[0118] Figure 4 A schematic diagram of a risk prediction method provided in an embodiment of the present application Figure 4 ,like Figure 4 As shown, in this embodiment Figure 1 Based on the embodiment, a possible implementation of the risk prediction method is described in detail, and the method includes:

[0119] S401. Acquire multiple sample data, each sample data carrying a corresponding timestamp.

[0120] In this embodiment, a large amount of sample data is collected through data collection means, and each sample data has a corresponding timestamp for marking the time point when the data is generated.

[0121] S402: Determine the ratio of sample data with risk events occurring to sample data with no risk events occurring according to labels in the plurality of sample data.

[0122] In this embodiment, the labels of all sample data are analyzed, and sample data with risk events and sample data without risk events are screened out according to the labels. The number of sample data with risk events and sample data without risk events are calculated respectively, and the ratio between the two is calculated.

[0123] S403: Determine the number of sample data with risk events occurring and the number of sample data with no risk events occurring that should be included in the training set, the validation set, and the test set, respectively, based on the quantity ratio.

[0124] In this embodiment, the distribution ratio of sample data with risk events and without risk events in the training set, validation set and test set is determined according to the obtained quantity ratio, and the number of sample data with risk events and the number of sample data without risk events that should be included in the training set, validation set and test set are calculated according to the determined distribution ratio and the total number of samples. By allocating according to the quantity ratio, each set can more realistically reflect the situation of risk events and non-risk events in the overall data, so that the risk prediction model can better cope with different situations during the training and testing process, and improve the generalization ability and reliability of the risk prediction model.

[0125] Especially for unbalanced sample data, this method can ensure that the ratio of the number of sample data with risk events and the number of sample data without risk events in each data set is relatively consistent, so as to improve the reliability of the risk prediction model.

[0126] S404. Allocate corresponding numbers of sample data with risk events occurring and sample data without risk events occurring to the training set, validation set, and test set in sequence according to the order of timestamps.

[0127] In this embodiment, according to the order of timestamps, starting from the earliest data, according to the previously determined quantity requirements, the corresponding number of sample data of risk events and risk events that have not occurred are allocated to the training set in turn; then the validation set is allocated; and finally the test set is allocated to ensure that the sample data in each set is also reasonably distributed in time sequence, which helps to more realistically simulate the actual situation.

[0128] A risk prediction method provided in an embodiment of the present application can ensure that the sample data in the training set, validation set and test set are continuous in time by allocating sample data in the order of timestamps, which helps the risk prediction model to better learn the time trend and pattern of the data. By allocating according to the quantity ratio, each set can more realistically reflect the risk events and non-risk events in the overall data, so that the risk prediction model can better cope with different situations during training and testing, improve the generalization ability and reliability of the risk prediction model, and further achieve the technical effect of improving the accuracy of risk prediction.

[0129] Figure 5 A schematic diagram of a risk prediction method provided in an embodiment of the present application Figure 5 ,like Figure 5 As shown, in this embodiment Figure 1 Based on the embodiment, a possible implementation method of processing the information to be processed based on multiple risk prediction models to obtain the predicted risk level output by each risk prediction model is described in detail. The method includes:

[0130] S501. Generate at least one type of derived characteristic information according to the type of characteristic information in the information to be processed and target business requirements.

[0131] In this embodiment, derived features, also called derived features or feature engineering, refer to new features created from the information to be processed, which can provide additional information about the data and help improve the performance of the risk prediction model.

[0132] Specifically, at least one kind of derived feature information is generated according to the type of feature information in the information to be processed and the target business requirements: if any feature information is time series data, at least one indicator feature corresponding to the time series data is extracted as the derived feature information; according to the target business requirements, at least two feature information are selected from the information to be processed, and composite features corresponding to the at least two feature information are generated as the derived feature information.

[0133] For example, derived feature information based on target business needs: for example, construct new features based on the characteristics of the quick loan business and risk assessment needs. For example, calculate the target object's debt repayment ratio, debt coverage ratio, liquidity ratio, etc. to measure its debt repayment ability; construct a consumption propensity index, etc. to measure the target object's consumption. Time series derived feature information: extract derived features from time series data such as transaction records and repayment records, such as the average transaction amount in the last N periods, the maximum number of overdue days, the proportion of overdue times, etc. These features can reflect the customer's recent behavior patterns and risk trends. According to the target business needs, cross-feature generation: generate new composite features through cross-combinations between features. For example, combine the customer's industry type with the regional risk index to generate an industry-region risk score; or combine the customer's profitability index with the debt burden index to measure its comprehensive debt repayment ability. These cross-features help capture the complex relationship between variables and improve the predictive ability of the risk prediction model. By generating a variety of appropriate derived feature information, it helps to improve the model's prediction and analysis capabilities, thereby improving the accuracy of the prediction.

[0134] S502: Input multiple types of feature information and at least one type of derived feature information of the target object into each risk prediction model to obtain the predicted risk level output by each risk prediction model.

[0135] In this embodiment, various types of feature information and derived feature information are input into each risk prediction model to obtain the predicted risk level output by each risk prediction model, which can provide more comprehensive and detailed information for each risk prediction model, thereby improving the accuracy of the risk prediction results.

[0136] The risk prediction method provided in the embodiment of the present application, by using various types of feature information and derived feature information as input to each risk prediction model, helps each risk prediction model to more accurately understand and analyze the status of the target object, thereby improving the accuracy of the risk prediction results.

[0137] In one possible implementation, Figure 6 As shown, Figure 6 A risk prediction method process diagram provided in the embodiment of the present application Figure 6 In this embodiment Figure 1 Based on the embodiment, a possible implementation of the risk prediction method is described in detail, and the method includes:

[0138] Obtain the information to be processed, perform feature selection on the information to be processed, and obtain the result of feature selection. Before obtaining the result of feature selection, perform baseline model processing and data exploration analysis on the result of feature selection; wherein, baseline model processing is used to evaluate the impact of feature selection on the performance of subsequent risk prediction models. The baseline model is trained using the results of all feature selections as a comparison object. By comparing the results of feature selection with the baseline model, it can be identified whether feature selection has brought performance improvement, and the extent of the improvement. This helps to judge the effectiveness of the feature selection strategy. Before feature selection, data exploration analysis can help understand the distribution, trend, and potential outliers of the data. This is very important for selecting appropriate features.

[0139] The result of feature selection is subjected to feature engineering processing to obtain target input data. The feature engineering processing includes data cleaning, feature construction and feature selection on the result of feature selection, wherein data cleaning refers to missing value processing, outlier processing and normalization processing on the result of feature selection. This method can not only improve the quality and reliability of the information to be processed, but also enhance the expression ability and prediction performance of the subsequent risk prediction model.

[0140] The target input data is respectively input into multiple risk prediction models to obtain the predicted risk level output by each risk prediction model, wherein the multiple risk prediction models include a random forest algorithm, an LSTM model and an XGBoot algorithm.

[0141] The predicted risk levels output by each risk prediction model are fused to obtain the fused predicted risk level, and the risk prediction result corresponding to the target object is determined based on the fused predicted risk level and risk threshold.

[0142] Figure 7 A schematic diagram of the structure of a risk prediction device provided in an embodiment of the present application is shown in FIG. Figure 7As shown, the risk prediction device 700 provided in this embodiment includes:

[0143] An acquisition module 701 is used to acquire information to be processed, where the information to be processed includes multiple types of feature information of a target object;

[0144] The first processing module 702 is used to process the information to be processed based on multiple risk prediction models respectively to obtain the predicted risk level output by each risk prediction model;

[0145] A reading module 703 is used to read the current weight distribution information and risk threshold from a preset storage space, wherein the weight distribution information is used to characterize the weight corresponding to each risk prediction model;

[0146] A fusion module 704 is used to perform weighted fusion on the predicted risk levels corresponding to the risk prediction models according to the weight distribution information to obtain a fused predicted risk level;

[0147] The second processing module 705 is used to determine the risk prediction result corresponding to the target object according to the fused predicted risk level and risk threshold.

[0148] In a possible implementation manner, the first processing module 702 is further configured to:

[0149] Obtain multiple sample data, each sample data carries a corresponding timestamp;

[0150] According to the labels and corresponding timestamps in the multiple sample data, the multiple sample data are divided into a training set, a validation set, and a test set; wherein the label is used to indicate whether a risk event occurs;

[0151] Multiple risk prediction models are trained, validated, and tested based on the training set, validation set, and test set.

[0152] In a possible implementation manner, the first processing module 702 is further configured to:

[0153] Determine, based on the labels in the plurality of sample data, the ratio of the number of sample data in which risk events have occurred to the number of sample data in which risk events have not occurred;

[0154] According to the quantity ratio, determine the number of sample data with risk events and the number of sample data without risk events that should be included in the training set, validation set, and test set respectively;

[0155] According to the order of timestamps, the corresponding number of sample data with risk events occurring and sample data without risk events occurring are allocated to the training set, validation set and test set respectively.

[0156] In a possible implementation manner, the first processing module 702 is further configured to:

[0157] generating at least one type of derived characteristic information according to the type of characteristic information in the information to be processed and the target business requirements;

[0158] Multiple types of feature information and at least one type of derived feature information of the target object are input into each risk prediction model to obtain the predicted risk level output by each risk prediction model.

[0159] In a possible implementation, the first fusion module 502 is further configured to:

[0160] If any feature information is time series data, at least one indicator feature corresponding to the time series data is extracted as derived feature information;

[0161] According to the target business requirements, at least two feature information are selected from the information to be processed, and composite features corresponding to the at least two feature information are generated as derived feature information.

[0162] In a possible implementation manner, the reading module 703 is further configured to:

[0163] At every preset period, verification information corresponding to multiple historical objects within the preset period is obtained, wherein the verification information corresponding to the historical objects includes labels corresponding to the historical objects and multiple types of feature information, and the labels are used to indicate whether a risk event occurs;

[0164] Based on the verification information corresponding to multiple historical objects within a preset period, multiple risk prediction models are verified to obtain the prediction accuracy corresponding to each risk prediction model, and the weight corresponding to each risk prediction model is updated according to the prediction accuracy corresponding to each risk model.

[0165] In a possible implementation manner, the reading module 703 is further configured to:

[0166] Verify each risk prediction model based on the verification information corresponding to multiple historical objects within a preset period to obtain the prediction accuracy corresponding to each risk prediction model;

[0167] According to the type of characteristic information in the verification information, the auxiliary evaluation index to be calculated is determined from a plurality of preset auxiliary evaluation indexes, wherein the plurality of auxiliary evaluation indexes include: an area under a receiver operating characteristic curve (ROC) and a discrimination evaluation index;

[0168] Calculate the auxiliary evaluation indicators corresponding to each risk prediction model based on the verification information corresponding to multiple historical objects within a preset period;

[0169] According to the prediction accuracy and auxiliary evaluation indicators corresponding to each risk prediction model, the weight corresponding to each risk prediction model is updated.

[0170] In a possible implementation manner, the reading module 703 is further configured to:

[0171] At every preset period, determine the risk events that occurred within the preset period, and update the risk threshold according to the determined risk events; and / or,

[0172] In response to a user changing risk preference information through an interactive interface, updating the risk threshold according to the risk preference information;

[0173] Among them, preference information is used to indicate the user's tolerance for risk.

[0174] In a possible implementation manner, the reading module 703 is further configured to:

[0175] Every preset period, determine the risk events that occur within the preset period;

[0176] According to the determined risk events, the performance indicators corresponding to multiple risk prediction models are calculated;

[0177] The risk threshold is updated according to the frequency of occurrence of risk events within a preset period and / or the performance indicators corresponding to multiple risk prediction models.

[0178] In a possible implementation manner, the reading module 703 is further configured to:

[0179] If the frequency of risk events in the preset period is lower than the frequency of risk events in the previous period, and the difference is greater than the preset frequency threshold, the risk threshold is lowered, and the degree of reduction is positively correlated with the difference;

[0180] If the frequency of risk events in the previous period is lower than the frequency of risk events in the preset period, and the difference is greater than the preset frequency threshold, the risk threshold will be raised, and the degree of increase is positively correlated with the difference.

[0181] In a possible implementation manner, the reading module 703 is further configured to:

[0182] According to the weights corresponding to the various risk prediction models, the performance indicators corresponding to the multiple risk prediction models are weighted and summed to obtain a comprehensive performance indicator;

[0183] The risk threshold is updated based on the comprehensive performance indicators.

[0184] In a possible implementation manner, the reading module 703 is further configured to:

[0185] If the comprehensive performance index corresponding to the preset period is lower than the comprehensive performance index of the previous period, and the difference is greater than the preset performance threshold, the risk threshold is lowered, and the degree of reduction is positively correlated with the difference;

[0186] If the comprehensive performance index corresponding to the previous period is lower than the performance index of the preset period, and the difference is greater than the preset performance threshold, the risk threshold is lowered, and the degree of reduction is positively correlated with the difference.

[0187] In a possible implementation manner, the reading module 703 is further configured to:

[0188] If the comprehensive performance index is greater than the comprehensive performance index threshold, the risk threshold is updated.

[0189] In a possible implementation manner, the reading module 703 is further configured to:

[0190] Update the time length corresponding to the preset cycle based on comprehensive performance indicators and the frequency of risk events;

[0191] Among them, the comprehensive performance index is positively correlated with the time length, and the occurrence frequency is negatively correlated with the time length.

[0192] A risk prediction device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be described in detail here.

[0193] Figure 8 This is a schematic diagram of the structure of a risk prediction device provided in an embodiment of the present application. Figure 8 As shown, the risk prediction device 800 provided in this embodiment includes: at least one processor 801 and a memory 802. Optionally, the device 800 also includes a communication component 803. The processor 801, the memory 802 and the communication component 803 are connected via a bus 804.

[0194] In a specific implementation process, at least one processor 801 executes the computer-executable instructions stored in the memory 802, so that at least one processor 801 executes the above method.

[0195] The specific implementation process of the processor 801 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0196] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the invention can be directly implemented as a hardware processor, or can be implemented by a combination of hardware and software modules in the processor.

[0197] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (NVM), such as at least one disk storage.

[0198] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of this application is not limited to only one bus or one type of bus.

[0199] An embodiment of the present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0200] An embodiment of the present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0201] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0202] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0203] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0204] It should be further noted that, although the various steps in the flowchart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowchart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0205] It should be noted that the terms "including" and "having" in the claims and the specification and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or are inherent to these processes, products or apparatuses.

[0206] It should be understood that the above-mentioned device embodiments are only illustrative, and the device of the present application can also be implemented in other ways. For example, the division of units / modules in the above-mentioned embodiments is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units, modules or components can be combined, or can be integrated into another system, or some features can be ignored or not executed.

[0207] In addition, unless otherwise specified, each functional unit / module in each embodiment of the present application may be integrated into one unit / module, each unit / module may exist physically separately, or two or more units / modules may be integrated together. The above-mentioned integrated unit / module may be implemented in the form of hardware or in the form of a software program module.

[0208] If the integrated unit / module is implemented in the form of hardware, the hardware may be a digital circuit, an analog circuit, etc. The physical implementation of the hardware structure includes but is not limited to transistors, memristors, etc. Unless otherwise specified, the processor may be any appropriate hardware processor, such as a CPU, a GPU, an FPGA, a DSP, an ASIC, etc. Unless otherwise specified, the storage unit may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory (RRAM), a dynamic random access memory (DRAM), a static random access memory (SRAM), an enhanced dynamic random access memory (EDRAM), a high-bandwidth memory (HBM), a hybrid memory cube (HMC), etc.

[0209] If the integrated unit / module is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a memory, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0210] In the above embodiments, the description of each embodiment has its own emphasis. For the part not described in detail in a certain embodiment, please refer to the relevant description of other embodiments. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0211] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0212] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A risk prediction method, characterized in that: include: Acquire information to be processed, where the information to be processed includes multiple types of feature information of the target object; Processing the information to be processed based on multiple risk prediction models respectively to obtain the predicted risk degree output by each risk prediction model; Reading current weight distribution information and risk threshold from a preset storage space, wherein the weight distribution information is used to characterize the weights corresponding to each risk prediction model; Performing weighted fusion of the predicted risk levels corresponding to the risk prediction models according to the weight distribution information to obtain a fused predicted risk level; A risk prediction result corresponding to the target object is determined according to the fused predicted risk level and the risk threshold.

2. The method according to claim 1, characterized in that The weight distribution information is determined in the following manner: At every preset period, obtaining verification information corresponding to a plurality of historical objects within the preset period, wherein the verification information corresponding to the historical objects includes a label corresponding to the historical objects and multiple types of feature information, wherein the label is used to indicate whether a risk event occurs; According to the verification information corresponding to the multiple historical objects within the preset period, the multiple risk prediction models are verified to obtain the prediction accuracy corresponding to each risk prediction model, and according to the prediction accuracy corresponding to each risk prediction model, the weight corresponding to each risk prediction model is updated.

3. The method according to claim 2, characterized in that According to the verification information corresponding to the multiple historical objects within the preset period, the multiple risk prediction models are verified to obtain the prediction accuracy corresponding to each risk prediction model, and according to the prediction accuracy corresponding to each risk prediction model, the weight corresponding to each risk prediction model is updated, including: Verifying each risk prediction model according to the verification information corresponding to the multiple historical objects within the preset period to obtain the prediction accuracy corresponding to each risk prediction model; According to the type of characteristic information in the verification information, determining the auxiliary evaluation index to be calculated from a plurality of preset auxiliary evaluation indicators, wherein the plurality of auxiliary evaluation indicators include: an area under a receiver operating characteristic curve (ROC) and a discrimination evaluation index; Calculate the auxiliary evaluation index corresponding to each risk prediction model according to the verification information corresponding to the multiple historical objects within the preset period; According to the prediction accuracy and auxiliary evaluation indicators corresponding to each risk prediction model, the weight corresponding to each risk prediction model is updated.

4. The method according to claim 1, characterized in that: The risk threshold is determined in the following manner: At every preset period, determine the risk events that occurred within the preset period, and update the risk threshold according to the determined risk events; and / or, In response to a user changing risk preference information through an interactive interface, updating the risk threshold according to the risk preference information; The preference information is used to indicate the user's tolerance for risk.

5. The method according to claim 4, characterized in that At every preset period, risk events occurring within the preset period are determined, and the risk threshold is updated according to the determined risk events, including: At every preset period, determining the risk events occurring within the preset period; Calculating performance indicators corresponding to the multiple risk prediction models according to the determined risk events; The risk threshold is updated according to the frequency of occurrence of risk events within the preset period and / or the performance indicators corresponding to the multiple risk prediction models.

6. The method according to claim 5, characterized in that The risk threshold is updated according to the frequency of occurrence of risk events within the preset period, including: If the frequency of occurrence of risk events in the preset period is lower than the frequency of occurrence of risk events in the previous period, and the difference is greater than the preset frequency threshold, the risk threshold is lowered, and the degree of reduction is positively correlated with the difference; If the frequency of occurrence of risk events in the previous period is lower than the frequency of occurrence of risk events in the preset period, and the difference is greater than the preset frequency threshold, the risk threshold is increased, and the degree of increase is positively correlated with the difference.

7. The method according to claim 5, characterized in that Updating the risk threshold according to the performance indicators corresponding to the multiple risk prediction models includes: According to the weights corresponding to the various risk prediction models, the performance indicators corresponding to the multiple risk prediction models are weighted and summed to obtain a comprehensive performance indicator; The risk threshold is updated according to the comprehensive performance indicator.

8. The method according to claim 7, characterized in that The risk threshold is updated according to the comprehensive performance indicator, including: If the comprehensive performance index corresponding to the preset period is lower than the comprehensive performance index of the previous period, and the difference is greater than the preset performance threshold, the risk threshold is lowered, and the degree of reduction is positively correlated with the difference; If the comprehensive performance index corresponding to the previous period is lower than the performance index of the preset period, and the difference is greater than the preset performance threshold, the risk threshold is lowered, and the degree of reduction is positively correlated with the difference.

9. The method according to claim 7, characterized in that: The risk threshold is updated according to the comprehensive performance indicator, including: If the comprehensive performance index is greater than the comprehensive performance index threshold, the risk threshold is updated.

10. The method according to claim 7, characterized in that Also includes: According to the comprehensive performance index and the frequency of occurrence of risk events, the time length corresponding to the preset period is updated; The comprehensive performance index is positively correlated with the time length, and the occurrence frequency is negatively correlated with the time length.

11. The method according to any one of claims 1 to 9, characterized in that: Before the information to be processed is processed respectively based on multiple risk prediction models, the method further includes: Obtain multiple sample data, each sample data carries a corresponding timestamp; According to the labels and corresponding timestamps in the multiple sample data, the multiple sample data are divided into a training set, a validation set and a test set; wherein the labels are used to indicate whether a risk event occurs; The multiple risk prediction models are trained, verified and tested based on the training set, validation set and test set.

12. The method according to claim 11, characterized in that Dividing the plurality of sample data into a training set, a validation set, and a test set according to the labels and corresponding timestamps in the plurality of sample data comprises: Determine, according to the labels in the plurality of sample data, a ratio of the number of sample data in which risk events have occurred to the number of sample data in which risk events have not occurred; According to the quantity ratio, determine the number of sample data with risk events occurring and the number of sample data with no risk events occurring that should be included in the training set, the validation set, and the test set respectively; According to the order of timestamps, corresponding numbers of sample data with risk events occurring and sample data without risk events occurring are allocated to the training set, validation set and test set in turn.

13. The method according to any one of claims 1 to 9, characterized in that: The information to be processed is processed based on multiple risk prediction models to obtain the predicted risk level output by each risk prediction model, including: generating at least one type of derived characteristic information according to the type of characteristic information in the information to be processed and target business requirements; Multiple types of feature information and at least one type of derived feature information of the target object are input into each risk prediction model to obtain the predicted risk level output by each risk prediction model.

14. The method according to claim 13, characterized in that At least one type of derived characteristic information is generated according to the type of characteristic information in the information to be processed and the target business requirement, including at least one of the following: If any feature information is time series data, extracting at least one indicator feature corresponding to the time series data as the derived feature information; According to the target business requirement, at least two feature information are selected from the information to be processed, and composite features corresponding to the at least two feature information are generated as derived feature information.

15. A risk prediction device, comprising: An acquisition module, used to acquire information to be processed, wherein the information to be processed includes multiple types of feature information of a target object; A first processing module, used to process the information to be processed based on multiple risk prediction models respectively, to obtain the predicted risk degree output by each risk prediction model; A reading module, used to read the current weight distribution information and risk threshold from a preset storage space, wherein the weight distribution information is used to characterize the weights corresponding to each risk prediction model; A fusion module is used to perform weighted fusion on the predicted risk levels corresponding to the risk prediction models according to the weight distribution information to obtain a fused predicted risk level; The second processing module is used to determine the risk prediction result corresponding to the target object according to the fused predicted risk level and the risk threshold.

16. A risk prediction device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 14.

17. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 14 when executed by a processor.

18. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 14 when being executed by a processor.

Images