Financial risk prediction method, prediction device and computer equipment

Abstract

The invention provides a financial risk prediction method and device and computer equipment, and belongs to the field of financial services, and the financial risk prediction method comprises the steps: obtaining a plurality of pieces of historical financial transaction data of a user, the financial transaction data comprising a credit record, a repayment record, a consumption record and an investment record; performing credit evaluation on the user based on the first credit evaluation model and the historical financial transaction data, and determining a predicted credit score; and importing the predicted credit score into a multi-task dichotomy prediction model for calculation to predict potential risk factors of the user in a financial transaction process, determining financial risks corresponding to the risk factors, and generating a personalized risk report of the user. According to the invention, customized financial suggestions can be provided for users, and the stability of financial transactions and the prediction precision of risks are improved.

Description

Technical Field

[0001] The present application relates to the field of financial services technology, and in particular to a financial risk prediction method, prediction device and computer equipment. Background Art

[0002] In the financial services industry, accurate prediction of financial risks is crucial for financial institutions and investors. Traditional financial risk assessment methods mainly rely on the experience of experts and static scoring models, which are often inadequate in dealing with complex and changing financial markets. With the advancement of big data and artificial intelligence technology, the use of machine learning and deep learning algorithms to predict financial risks has become a reality.

[0003] However, current financial risk prediction methods still face several problems. For example, many methods lack effective cross-validation techniques in feature selection and model training, which leads to insufficient generalization of the model. In addition, existing prediction models usually only focus on a single risk factor, while ignoring the interaction between multiple risk factors that may exist in financial transactions. Summary of the invention

[0004] The main purpose of the present invention is to provide a financial risk prediction method, aiming to improve the stability of financial transactions and the accuracy of risk prediction.

[0005] To achieve the above object, the present invention provides a financial risk prediction method, which comprises: Acquire multiple historical financial transaction data of the user, wherein the financial transaction data includes credit records, repayment records, consumption records, and investment records; Performing a credit assessment on the user based on the first credit assessment model and the historical financial transaction data to determine a predicted credit score; The predicted credit score is imported into a multi-task binary classification prediction model for calculation to predict the potential risk factors of the user in the financial transaction process, determine the financial risks corresponding to the risk factors, and generate a personalized risk report for the user.

[0006] Optionally, the step of performing credit assessment on the user based on the first credit assessment model and the financial transaction data to determine a predicted credit score includes: Acquire the user's current financial transaction data within a preset time period, wherein the current financial transaction data includes current credit application, current repayment behavior, and current consumption pattern; Analyzing the collected historical financial transaction records based on the first credit assessment model to extract key factors affecting the user's credit score, the key factors including the number of overdue repayments, credit limit utilization rate or consumption habits; Calculate a preliminary credit score for the user corresponding to the key factors and compare the score with industry standards or historical data to determine the relative position of the user's credit status; The user's relative credit position and current financial transaction data are combined to dynamically adjust the preliminary credit score and determine the user's predicted credit score.

[0007] Optionally, the step of analyzing the collected historical financial transaction records based on the first credit assessment model to extract key factors affecting the user's credit score includes: Data cleansing of historical financial transaction records to remove incomplete, erroneous or abnormal historical financial transaction records; Based on the statistical analysis method, identifying the financial transaction features corresponding to the user's credit score in the historical financial transaction records after data cleaning, the financial transaction features including repayment records, credit limit utilization rate, and number of overdue payments; Based on the machine learning algorithm, the financial transaction characteristics whose consumption records, browsing histories and behavioral habits meet the preset requirements are screened out; The screened financial transaction features are combined to generate multiple feature groups, and based on the cross-validation technology, the influence of each feature group on the credit score is evaluated to determine the key factors affecting the user's credit score.

[0008] Optionally, the step of combining the screened financial transaction features to generate a plurality of feature groups, and evaluating the degree of influence of each feature group on the credit score based on a cross-validation technique to determine the key factors affecting the user's credit score includes: Conduct stratified sampling on the screened financial transaction features so that each feature group contains features of different categories; Based on the cross-validation technology, the data set is divided into a training set and a validation set. Through multiple iterations, each feature group is tested on different data subsets, and the actual performance of each feature group in the validation set is recorded to determine the degree of influence of each feature group on the credit score, and generate corresponding evaluation results. The actual performance includes precision, recall rate and F1 score; Compare the evaluation results between the feature groups, identify the feature group that contributes most to the credit score, and determine that feature group as the key factor.

[0009] Optionally, the prediction method further includes: Obtain multiple historical credit data of the corresponding user; Preprocessing the historical credit data and constructing an initial credit assessment model related to the historical credit data; Based on a machine learning algorithm, current financial transaction data is imported into the initial credit assessment model for training to generate an intermediate credit assessment model; The intermediate credit assessment model is verified and tested, and an intermediate credit assessment model that meets the preset indicator performance is screened out, and the intermediate credit assessment model that meets the indicator performance is used as the first credit assessment model.

[0010] Optionally, the step of importing the predicted credit score into a multi-task binary classification prediction model for calculation to predict potential risk factors of the user in the financial transaction process, and determining the financial risks corresponding to the risk factors, and generating a personalized risk report for the user includes: The predicted credit score and the user's historical financial transaction data are input into a multi-task binary classification prediction model; In the multi-task binary classification prediction model, deep learning algorithms are used to extract features and recognize patterns from input data to identify potential risk factors; Matching the identified risk factors with the preset risk categories, determining the types of risks that the user may face during financial transactions, and generating corresponding matching results, which include credit risk, market risk or liquidity risk; Generate a personalized risk report corresponding to the matching result for the user; the risk report includes the risk type that the user may face, the risk level that the user may face, and corresponding prevention and response measures.

[0011] Optionally, the step of using a deep learning algorithm to perform feature extraction and pattern recognition on input data in the multi-task binary classification prediction model to identify potential risk factors includes: Perform preliminary feature extraction on the input data based on convolutional neural networks to capture the spatial correlation in financial transaction data; Processing time series data based on recurrent neural networks or long short-term memory networks to identify time-dependent features in financial transaction records; Based on the attention mechanism, the importance of different features is dynamically weighted to improve the sensitivity of the multi-task binary classification prediction model to key risk factors; Recognize the pattern and determine potential risk factors based on the spatial correlation in the financial transaction data, the time-dependent characteristics in the financial transaction records, and the multi-task binary classification prediction model with enhanced sensitivity; The step of matching the identified risk factors with the preset risk categories, determining the risk types that the user may face during the financial transaction process, and generating corresponding matching results includes: Constructing a mapping relationship database between risk factors and risk categories, wherein the mapping relationship database includes corresponding relationships between various risk factors and risk categories; Importing the identified risk factors into the mapping relationship database to determine the corresponding risk categories, and generating risk types that the user may face in the financial transaction process corresponding to the risk categories, wherein the risk types include credit risk, market risk, liquidity risk, operational risk and legal risk; The matched risk type is determined as the matching result.

[0012] Optionally, the step of obtaining a plurality of historical financial transaction data of the user includes: Based on the user's authorization permission, the user's transaction records, account information and credit history data are collected from multiple financial institutions and payment platforms to generate the historical financial transaction data.

[0013] In addition, to achieve the above-mentioned purpose, the present invention also provides a prediction device, which includes: a memory, a processor, and a financial risk prediction program stored in the memory and executable on the processor, wherein the financial risk prediction program is configured to implement the financial risk prediction method as described above.

[0014] In addition, to achieve the above object, the present invention also provides a computer device, including the prediction device as described above.

[0015] The embodiment of the present invention obtains multiple historical financial transaction data of a user, and performs credit assessment on the historical financial transaction data based on a first credit assessment model, thereby determining a predicted credit score, and importing it into a multi-task binary classification prediction model for calculation to predict the user's potential risk factors in the financial transaction process, and ultimately determines the financial risks corresponding to the risk factors, and generates a personalized risk report for the user, thereby providing the user with customized financial advice and improving the stability of financial transactions and the accuracy of risk prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0017] Figure 1 It is a flowchart of a financial risk prediction method according to an embodiment of the present invention; Figure 2 A schematic diagram of a financial risk prediction method flow chart according to another embodiment of the present invention; Figure 3 A schematic diagram of a financial risk prediction method according to another embodiment of the present invention; Figure 4 A schematic diagram of a financial risk prediction method according to another embodiment of the present invention; Figure 5 A schematic diagram of a financial risk prediction method according to another embodiment of the present invention; Figure 6 A schematic diagram of a financial risk prediction method flow chart according to another embodiment of the present invention; Figure 7 A schematic diagram of a financial risk prediction method according to another embodiment of the present invention; Figure 8 A schematic diagram of a financial risk prediction method according to another embodiment of the present invention; Fig. 9 The present invention is a flowchart of a financial risk prediction method according to another embodiment of the present invention.

[0018] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments, and the well-known modules, units and their connections, links, communications or operations are not shown or described in detail. In addition, the described features, architectures or functions can be combined in any way in one or more embodiments. It should be understood by those skilled in the art that the various embodiments described below are only for illustration and not for limiting the scope of protection of the present invention. It can also be easily understood that the modules or units or processing methods in the various embodiments described herein and shown in the drawings can be combined and designed according to various different configurations. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0020] The limitations of various nouns or methods referred to in the following embodiments, except for those that are logically untenable, are generally based on the broad concepts that can be implemented under the premise of the disclosure in the embodiments. Under such understanding, the various specific subordinate specific limitations of the nouns or methods should be regarded as the inventive content of the present invention, and should not be understood narrowly or interpreted in a biased manner on the grounds that the specification does not disclose the specific limitations. Similarly, under the premise that it can be implemented logically, the order of the steps in the method is flexible and changeable, and the specific subordinate specific limitations in the broad concepts of various nouns or methods all fall within the scope of protection of the present invention.

[0021] The main solution of the embodiment of the present application is: by obtaining multiple historical financial transaction data of the user, and performing credit assessment on the historical financial transaction data based on a first credit assessment model, thereby determining a predicted credit score, and importing it into a multi-task binary classification prediction model for calculation, so as to predict the user's potential risk factors in the financial transaction process, and finally determine the financial risks corresponding to the risk factors, and generate a personalized risk report for the user.

[0022] In this embodiment, for the convenience of description, the following description is made with the prediction device as the execution subject.

[0023] The existing financial risk prediction methods still face several problems. For example, many methods lack effective cross-validation techniques in the feature selection and model training stages, which leads to insufficient generalization of the models. In addition, existing prediction models usually only focus on a single risk factor, while ignoring the interaction between multiple risk factors that may exist in financial transactions.

[0024] This application provides a solution to provide users with customized financial advice, improve the stability of financial transactions and the accuracy of risk prediction.

[0025] To this end, the present invention proposes a financial risk prediction method; it can be understood that a prediction device for storing and executing the following method is provided in the computer device, and the prediction device can be implemented by a main controller, such as MCU (Microcontroller Unit), DSP (Digital Signal Process), FPGA (Field Programmable Gate Array), SOC (System On Chip), etc.

[0026] Reference Figure 1 In one embodiment of the present invention, the financial risk prediction method includes steps S100-S300, wherein: S100, obtaining multiple historical financial transaction data of the user, where the financial transaction data includes credit records, repayment records, consumption records, and investment records; S200, performing a credit assessment on the user based on the first credit assessment model and historical financial transaction data, and determining a predicted credit score; S300, importing the predicted credit score into the multi-task binary classification prediction model for calculation, so as to predict the potential risk factors of the user in the financial transaction process, determine the financial risks corresponding to the risk factors, and generate a personalized risk report for the user.

[0027] In this embodiment, the first credit assessment model can use machine learning algorithms, such as random forest, support vector machine or neural network, to improve the accuracy of credit scoring. By training these models, key factors affecting credit scoring, such as the user's income level, career stability, debt ratio and historical default record, can be identified.

[0028] In this embodiment, the multi-task binary classification prediction model can use deep learning technology, such as convolutional neural network (CNN) or recurrent neural network (RNN), to process and analyze large amounts of financial transaction data. These multi-task binary classification prediction models can capture complex patterns and time series features in the data, thereby more accurately predicting the risks that users may encounter during financial transactions.

[0029] In order to improve the accuracy of personalized risk reports, the prediction method can also use natural language processing (NLP) technology to analyze users' social media data, email communication records, and online behavior data. By analyzing these unstructured data, the prediction device can identify users' consumption habits, risk preferences, and potential fraudulent behavior tendencies, thereby providing users with a more comprehensive risk assessment.

[0030] In actual applications, the prediction device can also combine real-time data streams, such as market dynamics, economic indicators and industry news, to dynamically adjust credit scores and risk predictions. For example, when there are major market fluctuations or major events in the industry, the prediction device can instantly update the user's credit score and risk report, helping financial institutions to adjust credit policies and risk control strategies in a timely manner.

[0031] In actual applications, for example, if user A needs to conduct a credit financial transaction, after user A initiates a credit application, the prediction device will automatically call up its historical financial transaction data. These historical financial transaction data include user A's past credit records, repayment records, consumption records, and investment records in different financial institutions. Through these historical financial transaction data, the prediction device can fully understand user A's financial behavior and credit status.

[0032] Next, the prediction device will input these historical financial transaction data into the first credit assessment model. Assuming that we use the random forest algorithm, the model will identify the key factors affecting the credit score, such as income level, career stability, debt ratio, and historical default record, based on the historical financial transaction data of user A. Through the comprehensive evaluation of these factors, the prediction device can generate a predicted credit score. After obtaining the predicted credit score, the prediction device will import the score into the multi-task binary classification prediction model for analysis. Assuming that we use a recurrent neural network (RNN), this model can process and analyze a large amount of financial transaction data and capture complex patterns and time series characteristics in the data. In this way, the prediction device can predict the potential risk factors of user A in the financial transaction process and determine the financial risks corresponding to these risk factors.

[0033] In addition, in order to generate a more comprehensive personalized risk report, the prediction device can also use natural language processing (NLP) technology to analyze user A's social media data, email communication records, and online behavior data. Through the analysis of these unstructured data, the prediction device can identify user A's consumption habits, risk preferences, and potential fraudulent behavior tendencies. This information will be integrated into the risk report to provide financial institutions with more accurate risk assessments.

[0034] In addition, the prediction device can also combine real-time data streams, such as market dynamics, economic indicators and industry news, to dynamically adjust the credit score and risk forecast of user A. For example, if there are major fluctuations in the market or major events in the industry, the prediction device will instantly update the credit score and risk report of user A, helping financial institutions to adjust credit policies and risk control strategies in a timely manner.

[0035] In addition, in order to protect user privacy and data security, the prediction device can use advanced encryption technology and access control mechanisms. All user data will be encrypted during storage and transmission to ensure that only authorized financial institutions and individuals can access relevant information. At the same time, the prediction device will also conduct regular security audits and vulnerability scans to prevent data leaks and network attacks.

[0036] Further, see Figure 2 Another embodiment of the present invention provides a financial risk prediction method based on the above Figure 1 In the illustrated embodiment, the step of performing credit assessment on a user based on the first credit assessment model and the financial transaction data and determining a predicted credit score includes steps S210-S240, wherein: S210, obtaining the user's current financial transaction data within a preset time period, the current financial transaction data including current credit application, current repayment behavior and current consumption pattern; S220, analyzing the collected historical financial transaction records based on the first credit assessment model to extract key factors that affect the user's credit score, the key factors including the number of overdue repayments, the credit limit utilization rate, or consumption habits; S230, calculating a preliminary credit score for the user corresponding to the key factors, and comparing the score with industry standards or historical data to determine the relative position of the user's credit status; S240: Combining the relative position of the user's credit status and current financial transaction data to dynamically adjust the preliminary credit score and determine the user's predicted credit score.

[0037] In this embodiment, the first credit assessment model can use advanced machine learning algorithms to update and optimize model parameters in real time to adapt to the ever-changing financial environment. In order to improve the accuracy of credit scoring, the model can also combine the user's social network data and behavioral data, which help reveal the user's potential credit risk.

[0038] After acquiring the user's current financial transaction data, the prediction device will monitor and analyze this data in real time. For example, the prediction device will track the user's consumption behavior in real time and identify abnormal transaction patterns, such as sudden large-amount consumption or frequent small-amount transactions. These abnormal behaviors may have a negative impact on the user's credit score because they may indicate that the user is facing financial stress or is at risk of fraud.

[0039] The collected historical financial transaction records are analyzed through the first credit assessment model to determine the key factors that can affect the user's credit score. These key factors can be used to build a more accurate and personalized credit assessment system. Through in-depth research on these key factors, financial institutions can better understand the credit status of customers and provide them with more reasonable loan interest rates and credit limits. By calculating the preliminary credit scores corresponding to the key factors, financial institutions can better refine the credit rating and compare them with industry standards or historical data to divide customers into different credit rating groups. For example, customers with higher credit scores may be classified as "high-quality customers" and enjoy lower loan interest rates and higher credit limits. On the contrary, customers with lower credit scores may be classified as "risk customers", and financial institutions will adopt more cautious credit policies to reduce potential bad debt risks, which can better determine the relative position of users' credit status.

[0040] In addition, after the relative position of the user's credit status is determined, it can also be dynamically adjusted in combination with current financial transaction data to ensure the real-time and accuracy of the credit score. For example, if the user shows significant repayment ability and good consumption behavior in the short term, the prediction device can increase his credit score accordingly. On the contrary, if the user has overdue repayments or abnormal consumption patterns, the credit score will be reduced accordingly, thereby reminding financial institutions to adjust their credit policies in a timely manner.

[0041] In addition, financial institutions can also use these key factors to design more flexible financial products. For example, for customers with high credit scores but average income levels, financial institutions can provide some low-interest loan products with longer repayment periods to reduce customers' repayment pressure. For customers with low credit scores but stable income sources, financial institutions can design some short-term, high-interest loan products with smaller loan amounts to control risks.

[0042] To improve the accuracy of credit assessment, the prediction device can also integrate other data sources of the user, such as income level, occupational background, education level, etc. This information helps to build a more comprehensive user credit profile, thereby providing more reliable decision support for financial institutions. For example, a user with a stable income and a good education background may receive a higher score in the credit assessment because these factors are generally associated with a lower risk of default.

[0043] In addition, the prediction device can also use big data analysis technology to classify and cluster the credit behavior of different user groups. By identifying user groups with similar characteristics, financial institutions can better understand the credit risks of different customer groups and formulate more targeted marketing strategies and credit policies. For example, for those user groups with similar consumption habits and repayment behaviors, financial institutions can design specific financial products to meet their specific needs.

[0044] In practical applications, the prediction device can also provide real-time credit assessment reports to financial institutions through the user interface. These reports include not only the user's credit score and relative position of the credit status, but also the key factors affecting the credit score and the recommended credit policy. In this way, financial institutions can quickly respond to market changes and adjust credit strategies in a timely manner.

[0045] In addition, the prediction device will also combine the user's social network data to analyze the impact of their social relationships on the credit score. For example, if there are a large number of individuals with bad credit in the user's circle of friends, this may have a certain negative impact on the user's credit score. On the contrary, if the user maintains close contact with individuals with good credit, this may have a positive impact on the user's credit score.

[0046] Analysis of behavioral data is also an important part of credit assessment. Predictive devices collect user behavior data on the Internet, such as search history, browsing history, and online shopping habits. This data helps reveal the user's consumption preferences and potential risks. For example, users who frequently search for high-risk financial products or frequently visit gambling websites may be marked as high-risk users, thus affecting their credit scores.

[0047] Further, see Figure 3 Another embodiment of the present invention provides a financial risk prediction method based on the above Figure 2 In the illustrated embodiment, the steps of analyzing the collected historical financial transaction records based on the first credit assessment model and extracting the key factors affecting the user's credit score include steps S221-S224, wherein: S221. Data cleansing of historical financial transaction records to remove incomplete, erroneous or abnormal historical financial transaction records; S222. Based on a statistical analysis method, identify financial transaction features corresponding to the user's credit score in the historical financial transaction records after data cleaning, where the financial transaction features include repayment records, credit limit utilization rate, and number of overdue payments; S223. Based on the machine learning algorithm, screen out the financial transaction features whose consumption records, browsing histories and behavioral habits meet the preset requirements; S224. Combining the screened financial transaction features to generate multiple feature groups, and based on the cross-validation technology, evaluating the impact of each feature group on the credit score to determine the key factors affecting the user's credit score.

[0048] In this embodiment, the accuracy and reliability of the credit scoring model can be ensured by performing data cleaning on historical financial transaction records. Data cleaning is not limited to removing incomplete, erroneous or abnormal records, but also includes processing missing values. For those records that are missing key information, data can be supplemented by using interpolation, mean filling or model-based prediction methods to ensure the integrity of the data set.

[0049] Next, the cleaned data is mined using statistical analysis methods. By calculating the mean, median, standard deviation and other statistical indicators of various financial transaction characteristics, the prediction device can identify features that are highly correlated with the user's credit score. For example, the stability of repayment records, the volatility of credit limit utilization, and the frequency of overdue payments can largely reflect the user's credit status.

[0050] Based on machine learning algorithms, the prediction device can screen out financial transaction features that are closely related to credit scores. By using algorithms such as decision trees, random forests, and support vector machines, the prediction device can identify which consumption records, browsing histories, and behavioral habits have a significant impact on credit scores. These algorithms can help us screen out the most valuable features from a large number of features, thereby improving the prediction ability of the credit scoring model.

[0051] The screened financial transaction features are combined to generate multiple feature groups. These combinations include not only single features but also interactions between features. By constructing feature combinations, more complex credit risk patterns can be captured, thereby improving the model's predictive power.

[0052] Cross-validation techniques can be used to evaluate the degree of influence of each feature group on the credit score. Cross-validation techniques can effectively avoid overfitting and improve the generalization ability of the model by dividing the data set into multiple subsets and performing training and validation on different subsets. In this way, we can determine the key factors that affect the user's credit score, thereby providing a more accurate basis for the credit scoring model. The generated feature combinations are evaluated by using cross-validation techniques. The data set is divided multiple times on the training set to ensure that each subset has the opportunity to be used as a validation set, thereby reducing the risk of model overfitting. During the cross-validation process, the prediction accuracy and stability of each feature combination are recorded, and the best performing feature combination is finally selected.

[0053] Finally, the selected key factors are integrated into the credit scoring model. Through training and validation, the model is ensured to maintain high prediction accuracy and stability on different data sets. In addition, the model should also have good interpretability so that users and financial institutions can understand the key factors that affect credit scores and make more informed financial decisions.

[0054] In practical applications, this credit assessment model can provide financial institutions with more accurate credit scores, helping them reduce credit risks and improve operational efficiency in loan approval, credit card issuance and other businesses. At the same time, users can also understand their own credit status through the model and take corresponding measures to improve their credit scores, thereby obtaining better financial services.

[0055] Further, see Figure 4 Another embodiment of the present invention provides a financial risk prediction method based on the above Figure 3 In the embodiment shown, the steps of combining the screened financial transaction features to generate multiple feature groups, and evaluating the influence of each feature group on the credit score based on the cross-validation technology to determine the key factors affecting the user's credit score include steps S2241-S2243, wherein: S2241. Perform stratified sampling on the screened financial transaction features so that each feature group contains features of different categories; S2242. Based on the cross-validation technique, the data set is divided into a training set and a validation set. Through multiple iterations, each feature group is tested on different data subsets. The actual performance of each feature group in the validation set is recorded to determine the degree of influence of each feature group on the credit score, and the corresponding evaluation results are generated. The actual performance includes precision, recall and F1 score. S2243. Compare the evaluation results of each feature group, identify the feature group that contributes most to the credit score, and determine that feature group as the key factor.

[0056] In this embodiment, stratified sampling is performed on the screened financial transaction features to ensure that each feature group is representative in the data set. Stratified sampling not only helps to balance the distribution of various features, but also improves the generalization ability of the model. Next, the prediction device will use an advanced cross-validation technology - time series cross-validation to ensure that the data can also be effectively divided in the time dimension.

[0057] In time series cross validation, the dataset is divided into multiple consecutive subsets in chronological order. Each subset contains data within a certain time range to ensure that the model can learn the trends and cyclical changes in the time series. In this way, the performance of the credit scoring model in different time periods in actual applications can be simulated, thereby improving the robustness of the model.

[0058] When performing time series cross validation, the forecasting device will first select an initial time window as the training set and the remaining data as the validation set. Then the time window is gradually moved forward, one time unit at a time, and the training set and validation set are repartitioned. In this way, the forecasting device can obtain multiple combinations of training sets and validation sets, thereby performing multiple tests on each feature group.

[0059] In each iteration, the prediction device will record the actual performance of each feature group on the validation set, including accuracy, recall, and F1 score. These indicators will help us evaluate the contribution of each feature group to the credit score. In order to more comprehensively evaluate the performance of the model, the prediction device can also introduce other evaluation indicators, such as the area under the ROC curve (AUC) and the precision-recall curve (PR curve).

[0060] By comparing the evaluation results between the feature groups, the prediction device will identify the feature groups that contribute most to the credit score and determine them as key factors. These key feature groups will be further analyzed to reveal their specific impact mechanisms on the credit score. For example, some feature groups may perform well in predicting high-risk customers, while other feature groups may be more effective in identifying low-risk customers.

[0061] Further, see Figure 5 Another embodiment of the present invention provides a financial risk prediction method based on the above Figure 4 In the illustrated embodiment, the prediction method further includes steps S400-S700, wherein: S400, obtaining multiple historical credit data of the corresponding user; S500, preprocessing historical credit data and constructing an initial credit assessment model related to the historical credit data; S600, based on a machine learning algorithm, importing current financial transaction data into an initial credit assessment model for training to generate an intermediate credit assessment model; S700: Verify and test the intermediate credit assessment model, select the intermediate credit assessment model that meets the preset indicator performance, and use the intermediate credit assessment model that meets the indicator performance as the first credit assessment model.

[0062] In this embodiment, by preprocessing the historical credit data, outliers and noise data can be effectively removed to improve data quality. The preprocessing steps include data cleaning, data normalization, feature selection, etc. Data cleaning mainly removes duplicate records, fills missing values, and corrects erroneous data. Data normalization converts data of different dimensions and orders of magnitude to a unified scale to facilitate subsequent model training. Feature selection is to eliminate irrelevant or redundant features and retain features that have an important impact on credit assessment, thereby improving the generalization ability of the model.

[0063] When building the initial credit assessment model, a variety of statistical methods and machine learning algorithms can be used. Common statistical methods include logistic regression, linear discriminant analysis, etc., while machine learning algorithms include random forest, support vector machine, neural network, etc. The purpose of the initial model is to provide a basic framework for subsequent model training. Its performance does not need to meet the final requirements, but it needs to have a certain predictive ability.

[0064] In the process of importing current financial transaction data into the initial credit assessment model for training, methods such as cross-validation and grid search can be used to optimize parameters to improve the accuracy and stability of the model. Through continuous iteration and optimization, the intermediate credit assessment model gradually approaches the optimal solution, thereby generating a model that meets the preset indicator performance.

[0065] Verifying and testing the intermediate credit assessment model is a key step to ensure the reliability of the model. The verification process mainly evaluates the performance of the model on unseen data through methods such as cross-validation to ensure that the model has good generalization ability. The testing process is carried out on an independent test set to verify the predictive performance of the model. Only the intermediate credit assessment model that has passed verification and testing can be selected as the first credit assessment model for actual credit assessment tasks.

[0066] Further, see Figure 6 Another embodiment of the present invention provides a financial risk prediction method based on the above Figures 1 to 5 In any of the illustrated embodiments, the step of importing the predicted credit score into the multi-task binary classification prediction model for calculation to predict the potential risk factors of the user in the financial transaction process, and determining the financial risks corresponding to the risk factors, and generating a personalized risk report for the user includes steps S310-S340, wherein: S310, taking the predicted credit score and the user's historical financial transaction data as input, and inputting them into a multi-task binary classification prediction model; S320. In a multi-task binary classification prediction model, a deep learning algorithm is used to extract features and recognize patterns from input data to identify potential risk factors. S330, matching the identified risk factors with preset risk categories, determining the risk types that the user may face during financial transactions, and generating corresponding matching results, the matching results including credit risk, market risk or liquidity risk; S340. Generate a personalized risk report corresponding to the matching result for the user; the risk report includes the risk type that the user may face, the risk level that the user may face, and corresponding prevention and response measures.

[0067] In this embodiment, the predicted credit score and the user's historical financial transaction data are input into the multi-task binary classification prediction model, and the data is processed by the deep learning algorithm. The model can effectively extract key features and identify potential risk factors. These risk factors include but are not limited to overdue payment records, abnormal transaction behavior, asset volatility, etc.

[0068] After identifying these risk factors, the model matches them with pre-set risk categories. For example, if a user has multiple overdue payment records, the model may classify him as a high credit risk; if a user makes a large number of high-risk investments in a short period of time, the model may classify him as a high market risk; and if a user frequently makes large transactions in a short period of time, the model may classify him as a high liquidity risk.

[0069] The generated matching results will serve as a basis for generating a personalized risk report for the user. This report will not only list in detail the types of risks that the user may face, but also provide corresponding prevention and response measures based on the degree of risk. For example, for users with higher credit risk, the report may suggest that they improve their credit records, such as repaying on time and reducing outstanding debts; for users with higher market risk, the report may suggest that they diversify their investment portfolios to reduce the impact of fluctuations in a single asset; for users with higher liquidity risk, the report may suggest that they maintain a certain proportion of liquid funds to cope with sudden financial needs.

[0070] In addition, the prediction device can also provide customized financial product and service recommendations based on the user's personalized risk report. For example, for users with higher credit risk, financial institutions can provide credit repair services or low-risk financial products; for users with higher market risk, they can provide portfolio management services or market risk hedging tools; for users with higher liquidity risk, they can provide short-term financing solutions or liquidity management tools.

[0071] Through this multi-task binary classification prediction model, financial institutions can more accurately identify and manage the risks that users may face during financial transactions, thereby providing more personalized and targeted services and helping users better deal with various financial risks.

[0072] Further, see Figure 7 Another embodiment of the present invention provides a financial risk prediction method based on the above Figure 6 In the illustrated embodiment, in the multi-task binary classification prediction model, the step of using a deep learning algorithm to perform feature extraction and pattern recognition on input data to identify potential risk factors includes steps S321-S324, wherein: S321. Perform preliminary feature extraction on input data based on convolutional neural network to capture spatial correlation in financial transaction data. S322. Processing time series data based on recurrent neural networks or long short-term memory networks to identify time-dependent features in financial transaction records; S323. Based on the attention mechanism, the importance of different features is dynamically weighted to improve the sensitivity of the multi-task binary classification prediction model to key risk factors; S324. Based on the spatial correlation in the financial transaction data, the time-dependent characteristics in the financial transaction records, and the multi-task binary classification prediction model with improved sensitivity, pattern recognition is performed and potential risk factors are determined.

[0073] In this embodiment, a multi-level deep learning framework is constructed by comprehensively analyzing the spatial correlation in financial transaction data and the time-dependent features in financial transaction records to achieve comprehensive monitoring and prediction of financial risks. The framework first uses a convolutional neural network (CNN) to perform preliminary feature extraction on financial transaction data to capture the spatial correlation. CNN has achieved great success in the field of image recognition, and its powerful feature extraction capabilities are also applicable to financial data. Through the combination of convolutional layers, pooling layers, and activation functions, CNN can extract representative spatial features from raw data, laying the foundation for subsequent analysis.

[0074] Next, in order to process the time series information in financial transaction data, the framework uses recurrent neural networks (RNNs) and long short-term memory networks (LSTMs). RNNs can process sequence data and capture time-dependent features, but they are prone to gradient vanishing or exploding when processing long sequences. LSTM, as an improvement on RNN, effectively solves this problem by introducing a gating mechanism, allowing the model to learn long-term dependencies and better capture the time-dependent features in financial transaction records.

[0075] In order to improve the model's sensitivity to key risk factors, the framework introduces an attention mechanism. The attention mechanism can dynamically weight the importance of different features, allowing the model to pay more attention to those features that are critical to risk prediction when performing multi-task binary classification predictions. In this way, the model can more accurately identify potential risk factors and improve the accuracy and reliability of predictions.

[0076] After integrating spatial correlation, time-dependent features, and attention mechanisms, the framework performs pattern recognition on financial transaction data to identify potential risk factors. Through multi-level deep learning models, it is possible to conduct in-depth analysis of financial data from multiple dimensions, thereby achieving comprehensive monitoring and prediction of risks. Ultimately, the framework can provide financial institutions with more accurate risk assessment and early warning, help them formulate more effective risk management strategies, and ensure the stable operation of financial markets.

[0077] Further, see Figure 8 Another embodiment of the present invention provides a financial risk prediction method based on the above Figure 6 In the illustrated embodiment, the steps of matching the identified risk factors with the preset risk categories, determining the risk types that the user may face during the financial transaction process, and generating corresponding matching results include steps S331-S333, wherein: S331. Construct a mapping relationship database between risk factors and risk categories, wherein the mapping relationship database includes corresponding relationships between various risk factors and risk categories; S332, importing the identified risk factors into a mapping relationship database to determine the corresponding risk categories, and generating risk types that the user may face in the financial transaction process corresponding to the risk categories, the risk types including credit risk, market risk, liquidity risk, operational risk and legal risk; S333: Determine the matched risk type as the matching result.

[0078] In this embodiment, by constructing a mapping database, it is possible to quickly identify and classify potential risks in financial transactions. The mapping database is constructed based on a large amount of historical data and expert experience, ensuring that the correspondence between risk factors and risk categories has high accuracy and reliability.

[0079] In practical applications, it is necessary to first conduct a detailed analysis and identification of various risk factors that may arise in the financial transaction process. These risk factors may include but are not limited to market fluctuations, changes in credit ratings, counterparty performance capabilities, liquidity conditions, compliance of operating procedures, and changes in laws and regulations.

[0080] After the identified risk factors are imported into the mapping database, the prediction device will automatically match and analyze them to determine the corresponding risk category. For example, if the identified risk factors include a decline in the counterparty's credit rating, the prediction device will classify it as credit risk; if the risk factor involves market fluctuations that cause large fluctuations in asset prices, the prediction device will classify it as market risk.

[0081] After generating the risk types that users may face in the financial transaction process corresponding to the risk categories, the prediction device can also provide corresponding risk response measures and suggestions. For example, for credit risk, the prediction device can suggest that users take credit derivatives for hedging; for market risk, the prediction device can suggest that users diversify their investments to spread the risk.

[0082] Finally, the matched risk type is determined as the matching result and used as an important reference for financial transaction decision-making. In this way, financial institutions can manage and control risks more effectively and improve the security and stability of financial transactions.

[0083] Further, see Fig. 9 Another embodiment of the present invention provides a financial risk prediction method based on the above Figures 1 to 5 In any of the illustrated embodiments, the step of obtaining a plurality of historical financial transaction data of a user includes step S110, wherein: S110. Based on the user's authorization permission, collect the user's transaction records, account information, and credit history data from multiple financial institutions and payment platforms to generate historical financial transaction data.

[0084] In this embodiment, the user's financial data is integrated and analyzed through user authorization and permission, while ensuring the privacy of the user. First, the prediction device will require the user to perform a one-time authorization operation at various financial institutions and payment platforms to ensure the legitimacy of data acquisition. After the authorization is completed, the prediction device will automatically collect the user's transaction records, account information, and credit history data from various channels.

[0085] The collected data will be transmitted to a secure data center for de-identification to ensure that user privacy is not leaked. The de-identification process includes deleting or replacing all information that can directly identify the user, such as name, ID number, phone number, etc. The processed data will be stored in an encrypted database and can only be accessed by the forecasting device administrators and analysts authorized by the user.

[0086] The present invention also proposes a prediction device, which includes: the prediction device includes: a memory, a processor, and a financial risk prediction program stored in the memory and executable on the processor, the financial risk prediction program being configured to implement the above financial risk prediction method.

[0087] It is worth noting that since the prediction device of the present invention is based on the above-mentioned financial risk prediction method, the embodiments of the prediction device of the present invention include all technical solutions of all embodiments of the above-mentioned financial risk prediction method, and the technical effects achieved are also exactly the same, which will not be repeated here.

[0088] The present invention also provides a computer device, which includes the prediction device as described in the above embodiment.

[0089] It is worth noting that since the computer device of the present invention is based on the above-mentioned prediction device, the embodiments of the computer device of the present invention include all technical solutions of all embodiments of the above-mentioned prediction device, and the technical effects achieved are also exactly the same, which will not be repeated here.

[0090] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0091] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0092] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of various embodiments of the present invention.

[0093] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A financial risk prediction method, characterized in that: The financial risk prediction method comprises: Acquire multiple historical financial transaction data of the user, wherein the financial transaction data includes credit records, repayment records, consumption records, and investment records; Performing a credit assessment on the user based on the first credit assessment model and the historical financial transaction data to determine a predicted credit score; The predicted credit score is imported into a multi-task binary classification prediction model for calculation to predict the potential risk factors of the user in the financial transaction process, determine the financial risks corresponding to the risk factors, and generate a personalized risk report for the user.

2. The financial risk prediction method according to claim 1, characterized in that: The step of performing credit assessment on the user based on the first credit assessment model and the financial transaction data to determine a predicted credit score comprises: Acquire the user's current financial transaction data within a preset time period, wherein the current financial transaction data includes current credit application, current repayment behavior, and current consumption pattern; Analyzing the collected historical financial transaction records based on the first credit assessment model to extract key factors affecting the user's credit score, the key factors including the number of overdue repayments, credit limit utilization rate or consumption habits; Calculate a preliminary credit score for the user corresponding to the key factors and compare the score with industry standards or historical data to determine the relative position of the user's credit status; The user's relative credit position and current financial transaction data are combined to dynamically adjust the preliminary credit score and determine the user's predicted credit score.

3. The financial risk prediction method according to claim 2, characterized in that: The step of analyzing the collected historical financial transaction records based on the first credit assessment model to extract key factors affecting the user's credit score includes: Data cleansing of historical financial transaction records to remove incomplete, erroneous or abnormal historical financial transaction records; Based on the statistical analysis method, identifying the financial transaction features corresponding to the user's credit score in the historical financial transaction records after data cleaning, the financial transaction features including repayment records, credit limit utilization rate, and number of overdue payments; Based on the machine learning algorithm, the financial transaction characteristics whose consumption records, browsing histories and behavioral habits meet the preset requirements are screened out; The screened financial transaction features are combined to generate multiple feature groups, and based on the cross-validation technology, the influence of each feature group on the credit score is evaluated to determine the key factors affecting the user's credit score.

4. The financial risk prediction method according to claim 3, characterized in that: The step of combining the screened financial transaction features to generate multiple feature groups, and evaluating the influence of each feature group on the credit score based on the cross-validation technology to determine the key factors affecting the user's credit score includes: Conduct stratified sampling on the screened financial transaction features so that each feature group contains features of different categories; Based on the cross-validation technology, the data set is divided into a training set and a validation set. Through multiple iterations, each feature group is tested on different data subsets, and the actual performance of each feature group in the validation set is recorded to determine the degree of influence of each feature group on the credit score, and generate corresponding evaluation results. The actual performance includes precision, recall rate and F1 score; Compare the evaluation results between the feature groups, identify the feature group that contributes most to the credit score, and determine that feature group as the key factor.

5. The financial risk prediction method according to claim 4, characterized in that: The prediction method further comprises: Obtain multiple historical credit data of the corresponding user; Preprocessing the historical credit data and constructing an initial credit assessment model related to the historical credit data; Based on a machine learning algorithm, current financial transaction data is imported into the initial credit assessment model for training to generate an intermediate credit assessment model; The intermediate credit assessment model is verified and tested, and an intermediate credit assessment model that meets the preset indicator performance is screened out, and the intermediate credit assessment model that meets the indicator performance is used as the first credit assessment model.

6. The financial risk prediction method according to any one of claims 1 to 5, characterized in that: The step of importing the predicted credit score into a multi-task binary classification prediction model for calculation to predict the potential risk factors of the user in the financial transaction process, and determining the financial risks corresponding to the risk factors, and generating a personalized risk report for the user includes: The predicted credit score and the user's historical financial transaction data are input into a multi-task binary classification prediction model; In the multi-task binary classification prediction model, deep learning algorithms are used to extract features and recognize patterns from input data to identify potential risk factors; Matching the identified risk factors with the preset risk categories, determining the types of risks that the user may face during financial transactions, and generating corresponding matching results, which include credit risk, market risk or liquidity risk; Generate a personalized risk report corresponding to the matching result for the user; the risk report includes the risk type that the user may face, the risk level that the user may face, and corresponding prevention and response measures.

7. The financial risk prediction method according to claim 6, characterized in that: In the multi-task binary classification prediction model, the step of using a deep learning algorithm to extract features and recognize patterns on input data to identify potential risk factors includes: Perform preliminary feature extraction on the input data based on convolutional neural networks to capture the spatial correlation in financial transaction data; Processing time series data based on recurrent neural networks or long short-term memory networks to identify time-dependent features in financial transaction records; Based on the attention mechanism, the importance of different features is dynamically weighted to improve the sensitivity of the multi-task binary classification prediction model to key risk factors; Recognize the pattern and determine potential risk factors based on the spatial correlation in the financial transaction data, the time-dependent characteristics in the financial transaction records, and the multi-task binary classification prediction model with enhanced sensitivity; The step of matching the identified risk factors with the preset risk categories, determining the risk types that the user may face during the financial transaction process, and generating corresponding matching results includes: Constructing a mapping relationship database between risk factors and risk categories, wherein the mapping relationship database includes corresponding relationships between various risk factors and risk categories; Importing the identified risk factors into the mapping relationship database to determine the corresponding risk categories, and generating risk types that the user may face in the financial transaction process corresponding to the risk categories, wherein the risk types include credit risk, market risk, liquidity risk, operational risk and legal risk; The matched risk type is determined as the matching result.

8. The financial risk prediction method according to any one of claims 1 to 5, characterized in that: The step of obtaining a plurality of historical financial transaction data of a user comprises: Based on the user's authorization permission, the user's transaction records, account information and credit history data are collected from multiple financial institutions and payment platforms to generate the historical financial transaction data.

9. A prediction device, characterized in that: The prediction device includes: a memory, a processor, and a financial risk prediction program stored in the memory and executable on the processor, wherein the financial risk prediction program is configured to implement the financial risk prediction method according to any one of claims 1 to 8.

10. A computer device, characterized in that: Comprising the prediction device as claimed in claim 9.

Images