Risk assessment method and device, equipment, storage medium and computer program product

Abstract

The invention discloses a risk assessment method, device and equipment, a storage medium and a computer program product, and is applied to the technical field of artificial intelligence. The multi-dimensional data at least comprises historical loan data of the to-be-evaluated enterprise, operation data of the to-be-evaluated enterprise and macroeconomic indicators; carrying out data processing on the multi-dimensional data through a target model, and generating a risk probability containing to-be-evaluated enterprise default; according to the risk probability and a risk probability range of a plurality of risk levels corresponding to the business to be handled, determining a risk level corresponding to the enterprise to be assessed under the business to be handled, the business to be handled being a business requested to be handled by the enterprise to be assessed; and determining whether to handle the to-be-handled business for the to-be-assessed enterprise according to the Therefore, the scheme provided by the invention can improve the accuracy of enterprise default risk assessment.

Description

Technical Field

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

[0002] Small and micro enterprises are a general term for small enterprises, micro-enterprises, and family workshops. With the increase in the number of small and micro enterprises, it is of great significance to better serve them and handle various credit businesses for them.

[0003] Before applying for credit business for small and micro enterprises, it is usually necessary to predict the credit default risk of small and micro enterprises. However, small and micro enterprises usually lack sufficient credit history and financial transparency. Existing risk assessment methods usually only rely on outdated or incomplete data to conduct risk assessment on small and micro enterprises. This method cannot realize the ability to recognize complex patterns, and the results of risk assessment on small and micro enterprises are inaccurate. Summary of the invention

[0004] The embodiments of the present application provide a risk assessment method, apparatus, device, storage medium and computer program product, which can improve the accuracy of enterprise default risk assessment.

[0005] In a first aspect, an embodiment of the present application provides a risk assessment method, the method comprising: obtaining multi-dimensional data corresponding to an enterprise to be assessed, wherein the multi-dimensional data at least includes historical loan data of the enterprise to be assessed, operating data of the enterprise to be assessed, and macroeconomic indicators; performing data processing on the multi-dimensional data through a target model to generate a risk probability of default of the enterprise to be assessed, wherein the target model is a neural network model obtained by training a training set, and the training set at least includes multi-dimensional data of different enterprises and labels corresponding to different enterprises, and the labels are used to characterize the risk probability of default of the corresponding enterprise; determining the risk level corresponding to the enterprise to be assessed under the pending business according to the risk probability and the risk probability range of multiple risk levels corresponding to the pending business, wherein the pending business is the business requested by the enterprise to be assessed; determining whether to handle the pending business for the enterprise to be assessed according to the risk level.

[0006] In a second aspect, an embodiment of the present application provides a risk assessment device, which includes: a data acquisition module, used to acquire multi-dimensional data corresponding to the enterprise to be assessed, wherein the multi-dimensional data at least includes the historical loan data of the enterprise to be assessed, the operating data of the enterprise to be assessed, and macroeconomic indicators; a data processing module, used to process the multi-dimensional data through a target model to generate a risk probability of default of the enterprise to be assessed, wherein the target model is a neural network model obtained by training a training set, and the training set at least includes multi-dimensional data of different enterprises and labels corresponding to different enterprises, and the labels are used to characterize the risk probability of default of the corresponding enterprise; a level determination module, used to determine the risk level corresponding to the enterprise to be assessed under the pending business according to the risk probability and the risk probability range of multiple risk levels corresponding to the pending business, wherein the pending business is the business requested by the enterprise to be assessed; a business processing module, used to determine whether to handle the pending business for the enterprise to be assessed according to the risk level.

[0007] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, any one of the above-mentioned risk assessment methods is implemented.

[0008] In a fourth aspect, an embodiment of the present application provides a computer storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the risk assessment method described in the first aspect is implemented.

[0009] In a fifth aspect, an embodiment of the present application provides a computer program product. When instructions in the computer program product are executed by a processor of an electronic device, the electronic device is enabled to perform the risk assessment method described in the first aspect.

[0010] It can be seen from the above that in the embodiment of the present application, in the absence of credit history and financial transparency, the historical loan data, operating data and macroeconomic indicators of the enterprise are analyzed to assess the enterprise's default risk. Since the historical loan data, operating data and macroeconomic indicators of the enterprise can also reflect the repayment ability of the enterprise to a certain extent, the enterprise default risk determined by analyzing the historical loan data, operating data and macroeconomic indicators of the enterprise is more accurate than the prior art. Moreover, the target model is efficient and accurate in processing large-scale data and high-dimensional data. In the embodiment of the present application, the target model is used to analyze the multi-dimensional data of the enterprise, which can effectively improve the accuracy of the enterprise default risk assessment. Finally, in the embodiment of the present application, after obtaining the risk probability of enterprise default, the risk level of the enterprise is determined according to the risk probability, and the default risk of the enterprise is further refined, providing support for business decision-making.

[0011] It can be seen that the solution provided by the embodiment of the present application can improve the accuracy of enterprise default risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0013] Figure 1 is a flow chart of a risk assessment method provided by an embodiment of the present application;

[0014] Figure 2 is a flow chart of a method for training a target model provided by an embodiment of the present application;

[0015] Figure 3 It is a schematic diagram of a hyper-parameter tuning process of a target model provided by an embodiment of the present application;

[0016] Figure 4 is a schematic diagram of the structure of a risk assessment device provided by another embodiment of the present application;

[0017] Figure 5 It is a structural schematic diagram of an electronic device provided in yet another embodiment of the present application. DETAILED DESCRIPTION

[0018] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0019] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device 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 device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0020] It should be noted that the acquisition, storage, use and processing of data in the embodiments of the present application are in compliance with the relevant provisions of national laws and regulations.

[0021] In addition, it should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned, and they should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0022] For ease of understanding, before explaining the solution provided in the present application, the background of the solution provided in the present application is first explained.

[0023] In the field of financial risk management, credit scoring models, logistic regression, decision trees, etc., as well as machine learning and artificial intelligence technologies (e.g., neural networks, support vector machines) are usually used to assess the loan default risks of small and micro enterprises.

[0024] However, the existing method of assessing loan default risks of small and micro enterprises has the following defects:

[0025] (1) Model limitations: There are nonlinear relationships in the data related to small and micro enterprises and loans, and existing risk assessment models, such as logistic regression or simple decision trees, cannot effectively handle complex nonlinear relationships. Therefore, using this method to assess the loan default risk of small and micro enterprises has the problem of low assessment accuracy.

[0026] (2) Insufficient feature processing: When evaluating loan default risk of small and micro enterprises, some models may fail to fully utilize all available data features, or have deficiencies in feature selection and processing, thus affecting the accuracy of the evaluation results.

[0027] (3) Insufficient processing of big data: Data related to small and micro enterprises and loans contain high-dimensional data, and traditional models are inefficient in processing large-scale data sets, especially in high-dimensional data environments.

[0028] (4) Poor adaptability to new data: Existing technologies are not very adaptable when faced with new types of data or rapidly changing market conditions.

[0029] (5) Overfitting risk: Some models tend to overfit the training data, resulting in poor generalization ability on new data.

[0030] (6) Poor model stability: Some models exhibit unstable forecasting performance under different data sets or different market conditions.

[0031] (7) High computational cost: Some complex models may require high computing resources, which may not be economical or feasible in practical applications.

[0032] (8) Insufficient user-friendliness: Existing technologies may lack ease of use and require specialized knowledge to operate and maintain effectively.

[0033] (9) Poor adaptability to dynamic markets: Existing methods may not be able to effectively adapt to rapid changes in financial markets, such as economic fluctuations and policy changes.

[0034] In order to solve the problems of the prior art, embodiments of the present invention provide a risk assessment method, apparatus, device, storage medium and computer program product.

[0035] The following first introduces the risk assessment method provided by the embodiment of the present invention. The risk assessment method provided by the embodiment of the present application can be applied to a risk assessment platform.

[0036] Figure 1 FIG. 1 is a flow chart of a risk assessment method provided by an embodiment of the present invention. Figure 1 As shown, the method may include the following steps S101 to S104:

[0037] Step S101, obtaining multi-dimensional data corresponding to the enterprise to be evaluated.

[0038] In step S101, the enterprise to be evaluated is a small or micro enterprise, wherein the enterprise can be determined to be a small or micro enterprise based on indicators such as the number of employees, operating income, total assets, etc., combined with industry specifics.

[0039] In addition, in step S101, the multi-dimensional data at least includes historical loan data of the enterprise to be evaluated (for example, the amount of historical loans, the number of loans, and defaults), operating data of the enterprise to be evaluated (for example, sales data, financial data, customer data, market data, website data, social media data, user feedback data, brand evaluation data), and macroeconomic indicators (for example, gross domestic product, inflation rate, unemployment rate, international balance of payments, social consumption expenditure, fiscal revenue and expenditure, interest rate, exchange rate, money supply, deposit and loan balances of financial institutions, etc.). In the embodiment of the present application, the macroeconomic indicators may include the macroeconomic indicators of the enterprise to be evaluated during the historical loans and the macroeconomic indicators corresponding to the current period.

[0040] In one embodiment, the risk assessment platform can obtain data of different dimensions through different channels. For example, the historical loan data of the enterprise to be assessed can be obtained through a credit inquiry platform or a financial institution; the operation data of the enterprise can be provided by the enterprise or obtained through Internet inquiries; macroeconomic indicators can be obtained through official documents from the government or through news or reports on the economy published on the Internet.

[0041] Step S102, processing the multi-dimensional data through the target model to generate a risk probability of default of the enterprise to be evaluated.

[0042] In step S102, the target model may be a random forest model, which is an ensemble learning method that improves the accuracy and stability of the model by combining the prediction results of multiple decision trees. Ensemble learning is a technique that combines weak learners (such as decision trees) into a strong learner. The decision tree is the basic building block of the random forest model, which simulates the decision-making process from root to leaf in a tree structure.

[0043] In step S102, the target model is a neural network model obtained by training a training set, where the training set includes at least multi-dimensional data of different enterprises and labels corresponding to different enterprises, where the labels are used to characterize the risk probability of default of the corresponding enterprise.

[0044] It should be noted that the random forest model is efficient and accurate in processing large-scale data and high-dimensional data. In the embodiment of the present application, the random forest model is used to analyze the multi-dimensional data of the enterprise, which can effectively improve the accuracy of the enterprise default risk assessment.

[0045] Step S103, determining the risk level corresponding to the enterprise to be assessed under the pending business according to the risk probability and the risk probability ranges of the multiple risk levels corresponding to the pending business.

[0046] In step S103, the pending business is the business requested by the enterprise to be evaluated.

[0047] In an example, different businesses may correspond to different numbers of risk levels, and the risk probability range corresponding to each risk level may also be different. For example, for business A, it corresponds to four risk levels, and the risk probability ranges corresponding to the four risk levels are 0% to 30%, 30% to 50%, 50% to 80%, and 80% to 100%, respectively; for business B, it corresponds to three risk levels, and the risk probability ranges corresponding to the three risk levels are 0% to 60%, 60% to 80%, and 80% to 100%, respectively.

[0048] Step S104, determining whether to process pending business for the enterprise to be assessed based on the risk level.

[0049] In one example, if the risk level is low, for example, the risk level is lower than the preset level, the risk assessment platform determines that the enterprise to be assessed is a low-risk enterprise and can process a loan for it. In addition, when processing a loan for it, more favorable conditions can be provided to the enterprise to be assessed, for example, exemptions for the first few repayment cycles of the loan, or low interest rate discounts, etc.; if the risk level is high, for example, the risk level is higher than the preset level, the risk assessment platform determines that the enterprise to be assessed is a high-risk enterprise and prohibits it from processing a loan, or, when processing a loan, takes more stringent credit measures.

[0050] Based on the scheme defined by the above steps S101 to S104, it can be known that in the embodiment of the present application, in the absence of credit history and financial transparency, the historical loan data, operating data and macroeconomic indicators of the enterprise are analyzed to assess the default risk of the enterprise. Since the historical loan data, operating data and macroeconomic indicators of the enterprise can also reflect the repayment ability of the enterprise to a certain extent, the enterprise default risk determined by analyzing the historical loan data, operating data and macroeconomic indicators of the enterprise is more accurate than the prior art. Moreover, the target model is efficient and accurate in processing large-scale data and high-dimensional data. In the embodiment of the present application, the target model is used to analyze the multi-dimensional data of the enterprise, which can effectively improve the accuracy of the enterprise default risk assessment. Finally, in the embodiment of the present application, after obtaining the risk probability of enterprise default, the risk level of the enterprise is determined according to the risk probability, and the default risk of the enterprise is further refined, providing support for business decision-making.

[0051] It can be seen that the solution provided by the embodiment of the present application can improve the accuracy of enterprise default risk assessment.

[0052] The following is an introduction to the implementation process of the method provided in the embodiment of the present application.

[0053] In the embodiment of the present application, the target model is used to evaluate the multi-dimensional data of the enterprise to be evaluated to determine the risk level of the enterprise to be evaluated, wherein the target model can be used as follows: Figure 2 The model training method shown is used for training. Figure 2 As shown, the training process of the target model may include the following steps:

[0054] Step S201, obtaining data samples.

[0055] In step S201, a data sample can be constructed from multi-dimensional data of enterprises with loan behaviors, and the data sample includes at least multi-dimensional data corresponding to multiple enterprises and risk probabilities corresponding to multiple enterprises, wherein the risk probability is a label corresponding to the data feature, and the data feature is a feature obtained by extracting features from the multi-dimensional data.

[0056] Step S202: randomly divide the data samples according to a preset ratio to obtain a training set and a test set.

[0057] Step S203: extract data features corresponding to the multi-dimensional data and labels corresponding to the data features from the training set.

[0058] Step S204: construct a classifier, and initialize the classifier using a preset number of decision trees and a preset decision tree depth to obtain an initialized classifier.

[0059] In step S204, the target model is a random forest model, and accordingly, the classifier may be a random forest classifier.

[0060] Step S205 , inputting data features corresponding to the multi-dimensional data and labels corresponding to the data features into the initialized classifier to train the target model.

[0061] After obtaining the data sample, in order to make the data sample better for model training and thus improve the evaluation accuracy of the target model, the data sample needs to be preprocessed. In the embodiment of the present application, the data sample can be preprocessed by any one or more of missing value processing, outlier processing, feature encoding technology, and data standardization processing.

[0062] For missing values, in an embodiment of the present application, fields with missing values ​​are identified from the data sample, the values ​​of the fields without missing values ​​are statistically processed, and the statistical values ​​are filled into the fields with missing values.

[0063] It should be noted that in the above embodiments, the statistical processing performed on the fields with non-missing values ​​can be one or more of the mean, median, mode, and front and back data filling processing methods. The data type corresponding to the field with missing values ​​can be a numeric type or a categorical field. Taking the mean filling method as an example, the average value of each field is calculated, and the fields with missing values ​​are filled with these means to maintain the overall consistency of the data. This process can be implemented with the help of the Pandas library. For example, the missing values ​​in the specified column are filled by the Pandas library, and the filled value is the average value of the column.

[0064] By filling in missing fields with values, we can ensure data integrity and provide a basis for improving the accuracy of model output results.

[0065] For outliers, a box plot is performed on the data sample to identify the outliers in the data sample, and the outliers are processed with standard scores.

[0066] Specifically, in the embodiment of the present application, a box plot and a statistical analysis method (e.g., a Z-score method) are used to identify and process outliers to improve data quality. For example, outliers are identified through a box plot, and these outliers are processed using a Z-score method, such as by calculating the standard deviation of each data point from the mean, and treating data points with a deviation exceeding a certain threshold (e.g., two standard deviations) as outliers and processing them accordingly.

[0067] For feature encoding, in the embodiment of the present application, non-numeric fields are identified from the data sample, and the data of the non-numeric fields are converted to numeric types. As an example, for non-numeric fields, one-hot encoding or label encoding can be used for feature conversion, for example, through the `OneHotEncoder` in Scikit-learn to achieve feature conversion of non-numeric fields, converting categorical data into a multi-dimensional representation of numeric type, which is convenient for model processing.

[0068] By using one-hot encoding or label encoding for non-numeric fields, the model can effectively process this information.

[0069] For data standardization, in the embodiment of the present application, the numerical features in the data sample are standardized, that is, the numerical features are processed using the standardization (StandardScaler) or normalization (MinMaxScaler) technology to improve the efficiency and stability of model training. For example, a standardization method is used to scale each feature so that its mean is 0 and its standard deviation is 1. This method can be implemented through the `StandardScaler` class in the Scikit-learn library to effectively improve the model's efficiency in interpreting data and the accuracy of prediction.

[0070] By conducting comprehensive data preprocessing on data samples, a solid foundation is laid for subsequent model training and analysis.

[0071] Furthermore, after preprocessing the data samples, the data samples can be divided into a test set and a training set, wherein the training set is used for training the target model and the test set is used for performance testing of the trained target model.

[0072] As an example, the train_test_split method of Scikit-learn can be used to split the data set to ensure a reasonable distribution between the training set and the test set, thereby ensuring the effectiveness and accuracy of the model evaluation. Among them, the division ratio between the training set and the test set (i.e., the preset ratio) can be a fixed value or adjusted according to actual needs. For example, the preset ratio test_size can be set to 0.3, that is, 30% of the data sample is randomly selected as the test set, and the remaining 70% is used as the training set.

[0073] To implement this process, you first need to import the training set split function train_test_split from the Scikit-learn library. Then, pass the complete dataset and its corresponding labels as input parameters to the training set split function train_test_split. By setting test_size = 0.3, the training set split function train_test_split automatically randomly allocates data samples into training and test sets to ensure the independence and representativeness of the two datasets. This data splitting method helps to evaluate the performance of the model on unknown data and can reduce the risk of overfitting.

[0074] In addition, in the above embodiment, in the process of dividing the data samples according to the preset ratio, a random division method is adopted to ensure the repeatability of the experiment, so that the data division method remains consistent each time the code is run. This method can ensure that the target model has sufficient data support during the training and testing process.

[0075] Furthermore, after completing the data sample division, Scikit-learn's train_test_split can be used to divide the training and test sets. Next, the target model is built and optimized, including parameter initialization, training, and hyperparameter tuning. Finally, the model is evaluated through a variety of indicators, and predict_proba and feature_importances_ are used to predict new data and analyze feature impact.

[0076] In the present application embodiment, building a target model is one of the core steps. First, the classifier is initialized by the Scikit-learn library. In this initialization process, two key hyperparameters are set: the number of decision trees (n_estimators) is set to 100, and the maximum depth of the decision tree (max_depth) is set to 5. The initial setting of these parameters is intended to provide a reasonable starting point for subsequent tuning work.

[0077] After the model is initialized, the model is trained using the training set data. This is done by calling the `fit` method of the target model, which takes the features and labels of the training set as input. During the training process, the target model will learn how to predict labels based on the provided features, thus preparing for subsequent prediction tasks.

[0078] After the target model is obtained through training, in order to further improve the performance of the model, in the embodiment of the present application, the target model is subjected to hyper-parameter tuning, specifically, Figure 3 As shown, this process can be achieved through the following steps:

[0079] Step S1, performing an accuracy test on the output result of the target model using a test set to obtain a first prediction accuracy;

[0080] Step S2, detecting whether the first prediction accuracy is less than a preset accuracy.

[0081] Step S3, when the first prediction accuracy is less than the preset accuracy, adjusting the model parameters of the target model to obtain an adjusted target model, wherein the model parameters include the number of decision trees and / or the depth of the decision trees;

[0082] Step S4, perform accuracy test on the output result of the adjusted target model through the test set to obtain a second prediction accuracy, and update the first prediction accuracy to the second prediction accuracy, and repeat steps S2 to S4 until the first prediction accuracy is less than the preset accuracy.

[0083] In the above embodiment, GridSearchCV or RandomizedSearchCV can be used for hyperparameter tuning to determine the optimal model configuration parameters by systematically testing different parameter combinations. For example, by setting different combinations of the number of decision trees and the maximum depth, to determine which combination can provide the best prediction performance. The tuning process is iterative and may require multiple runs to find the most appropriate parameter settings.

[0084] As an example, the number of trees in the forest is set to 100, the maximum depth is 5, and the control parameter is 42. The classifier is initialized with these parameters. Then, GridSearchCV is set for parameter tuning, and the number of decision trees and the maximum depth are set to (100,5), (200,10), (300,15). Then a GridSearchCV instance is created and the training data is used for parameter tuning.

[0085] Through the above process, not only can the basic target model be constructed, but also the target model can be ensured to achieve the optimal prediction performance through fine parameter adjustment.

[0086] In the embodiment of the present application, after obtaining the target model, the model evaluation phase is entered, and multiple indicators are used to comprehensively evaluate the performance of the target model. Among them, the performance indicators for evaluating the target model may include but are not limited to: accuracy, recall, F1 score, and the value of the area under the receiver operating characteristic curve (AUC). In addition, a confusion matrix can also be used to display the details of the target model prediction, including the number of true categories, false positive categories, true negative categories, and false negative categories.

[0087] For accuracy, the performance of the target model on the test set is measured by the accuracy metric, which reflects the proportion of correct predictions made by the target model. Recall measures the proportion of positive categories identified by the target model to the total number of actual positive categories, while the F1 score is the harmonic mean of precision and recall, which is used to measure the overall performance of the target model.

[0088] Next, the AUC value provides an evaluation of the performance of the target model at different thresholds; the confusion matrix provides an intuitive way to view the performance of the model on each category, including its predictive ability for positive and negative classes.

[0089] By comprehensively evaluating the target model, we can accurately understand its performance in actual scenarios, thus providing a solid foundation for further optimization and application.

[0090] In order to enhance the interpretability of the target model, in an embodiment of the present application, after the target model is trained, the contribution of each feature to the evaluation accuracy of the target model can also be analyzed.

[0091] Specifically, the output result of the target model is tested for accuracy through a test set to obtain a first prediction accuracy; then, the data features in the test set are adjusted to obtain an adjusted test set; the output result of the target model is tested for accuracy through the adjusted test set to obtain a third prediction accuracy; based on the difference between the first prediction accuracy and the third prediction accuracy, the feature importance of the adjusted data features is determined; when the feature importance is greater than the preset importance, the adjusted data features are determined to be key input data of the target model.

[0092] In the above embodiment, the feature importance is used to characterize the influence of the adjusted data feature on the prediction accuracy of the target model.

[0093] In the application and interpretation phase, the `predict_proba` method of the trained target model is used to predict the default probability of new data. This method not only gives the final classification results, but also provides the probability estimates belonging to each category, making the evaluation results richer and more interpretable. This method allows decision makers to make more flexible decisions based on probability thresholds, which is particularly important for risk management.

[0094] Secondly, to enhance the interpretability of the target model, in the embodiment of the present application, the feature_importances attribute of the target model is used to analyze the impact of each data feature on the prediction of the target model. This attribute provides a feature importance score, which shows the relative importance of each data feature in the decision-making process of the target model. By analyzing the feature importance corresponding to these data features, it is possible to understand which data features have a greater impact on the evaluation results, thereby providing guidance for data feature selection and model optimization.

[0095] In this step, the new data is first predicted using the predict_proba method of the trained target model to obtain an evaluation result on the probability of default. The predict_proba method provides an estimate of the probability of belonging to each category (here, default and non-default) for each case. These evaluation results are then converted into a structured data table, which usually includes an identifier for each case (for example, a corporate loan account) and the corresponding probability of default.

[0096] Next, the generated data table is exported to a standard format file, such as CSV, for further analysis or reporting. The export process can be implemented through the corresponding data processing library (Pandas library) in the programming language. Finally, the exported file will be uploaded or imported into a data warehouse for long-term storage and management. The information in the data warehouse can be used for subsequent data analysis, business decision support, and credit strategy adjustment.

[0097] It should be noted that the above process not only ensures the accuracy and real-time nature of the prediction results, but also provides support for a wider range of business intelligence and data analysis activities by integrating the data into the data warehouse.

[0098] In the subsequent steps of this technical solution, we focus on the continuous maintenance of the model and the implementation of business strategies. First, in order to maintain the accuracy and timeliness of the model, we regularly update and retrain the model with newly collected data. This process includes collecting the latest data, re-performing the data preprocessing steps, and retraining the model with these updated data sets. Regularly updating the model ensures that it can capture the latest trends and changes in the market and industry environment, thereby improving its effectiveness in real-world scenarios.

[0099] When the target model is tested using the test set, and the test results indicate that the target model can be applied in practice, the risk assessment platform uses the target model to assess the default risk of the enterprise to be assessed, outputs the probability of default, and determines the risk level corresponding to the enterprise to be assessed. Then, it is determined whether to handle pending business for the enterprise to be assessed based on the risk level. In particular, if the risk level is higher than the preset level, it is prohibited to handle pending business for the enterprise to be assessed. In the case where the risk level is lower than the preset level, it is allowed to handle pending business for the enterprise to be assessed, and the business handling strategy for the enterprise to be assessed to handle pending business is determined based on the level corresponding to the risk level.

[0100] For example, when the risk level of the enterprise to be assessed is low, preferential business conditions can be provided; when the risk level of the enterprise to be assessed is high, stricter credit measures may need to be taken.

[0101] In one embodiment, when there are multiple companies to be evaluated, risk level distribution information of risk levels corresponding to the multiple companies to be evaluated can also be obtained, and the business handling strategies for pending businesses at different risk levels can be adjusted according to the risk level distribution information.

[0102] In the above embodiment, the business handling strategy can be formulated or adjusted in combination with the business knowledge and the prediction results of the model. This includes analyzing the key risk factors identified by the model, evaluating the probability of default of different customer groups, and adjusting the credit limit, interest rate or other relevant credit conditions accordingly. For example, for customer groups with a higher predicted probability of default, stricter credit measures are taken, while more favorable conditions can be provided for customers with lower risks; for another example, for a certain type of enterprise, its probability of default is high, then it can be determined that the business handling strategy is set unreasonably, such as the repayment period is set too short, or the repayment interest rate is set too high, etc. In this scenario, the repayment period or the repayment interest rate can be adjusted.

[0103] The above process not only enhances the effectiveness of the target model in practical applications, but also ensures that the output of the target model can be actually reflected in business decisions, thereby improving the efficiency of corporate credit management while reducing potential credit risks. Through such continuous maintenance and strategy adjustments, the model can better serve business needs and realize data-driven decision-making processes.

[0104] This concludes the introduction to the method provided in the embodiments of the present application.

[0105] From the above introduction, it can be seen that the method provided by the embodiment of the present application has at least the following advantages:

[0106] (1) Improving the prediction accuracy of default risk: The method provided in the embodiments of the present application utilizes the powerful data processing capabilities of random forests and their high sensitivity to complex patterns, which can effectively improve the accuracy of default probability prediction. By integrating multiple decision trees, the risk of overfitting is reduced, thereby achieving more reliable prediction results on a variety of data sets.

[0107] (2) Optimize risk management: With more accurate predictions of default probability, financial institutions can better manage and mitigate credit risk. The solution supports data-driven adjustments to credit strategies, thereby optimizing asset allocation while controlling risk.

[0108] (3) Enhance model interpretability and transparency: Feature importance analysis provides decision makers with a transparent view of the internal operation of the model, increases the interpretability of the model, and helps build user trust in the model. Easy-to-interpret model outputs promote cross-departmental collaboration and enable non-technical personnel to understand and use prediction results.

[0109] (4) Improve data processing efficiency: Through the systematic data preprocessing steps, missing values ​​and outliers are effectively handled, ensuring data quality, thereby improving the performance of the entire model. The effective division of the data set ensures the data validity and representativeness of the model during the training and testing phases.

[0110] (5) Adaptability and flexibility: Regularly updating the model to adapt to new data ensures that the model remains current and relevant as the market and environment change. The flexible model structure allows adjustment and optimization according to business needs, providing a wider range of application possibilities.

[0111] (6) Long-term business value: By integrating into the data warehouse, the forecast results can be used for long-term strategic planning and business decision-making. This promotes the data-based decision-making process and improves the company's risk assessment capabilities and market competitiveness in the long term.

[0112] Based on the risk assessment method provided in the above embodiment, the present application also provides a specific implementation of the risk assessment device. Please refer to the following embodiment.

[0113] See first Figure 4 The risk assessment device 400 provided in the embodiment of the present application includes: a data acquisition module 401, a data processing module 402, a level determination module 403 and a business handling module 404.

[0114] The data acquisition module 401 is used to acquire multi-dimensional data corresponding to the enterprise to be evaluated, wherein the multi-dimensional data at least includes historical loan data, operation data and macroeconomic indicators of the enterprise to be evaluated;

[0115] The data processing module 402 is used to process the multi-dimensional data through the target model to generate a risk probability of default of the enterprise to be evaluated, wherein the target model is a neural network model obtained by training the training set, and the training set at least includes the multi-dimensional data of different enterprises and the labels corresponding to the different enterprises, and the labels are used to characterize the risk probability of default of the corresponding enterprise;

[0116] The level determination module 403 is used to determine the risk level corresponding to the enterprise to be assessed under the pending business according to the risk probability and the risk probability range of multiple risk levels corresponding to the pending business, wherein the pending business is the business requested by the enterprise to be assessed;

[0117] The business processing module 404 is used to determine whether to process pending business for the enterprise to be assessed according to the risk level.

[0118] In one embodiment, the risk assessment device includes a model training module, which is used to train a target model through the following steps: obtaining a data sample, wherein the data sample includes at least multi-dimensional data corresponding to multiple enterprises and risk probabilities corresponding to multiple enterprises; randomly dividing the data sample according to a preset ratio to obtain a training set and a test set; extracting data features corresponding to the multi-dimensional data and labels corresponding to the data features from the training set; constructing a classifier, and initializing the classifier with a preset number of decision trees and decision tree depth to obtain an initialized classifier; inputting the data features corresponding to the multi-dimensional data and the labels corresponding to the data features into the initialized classifier to train the target model.

[0119] In one embodiment, the model training module includes: a preprocessing module, which is used to preprocess the data samples after obtaining the data samples in any one or more of the following ways: identifying fields with missing values ​​from the data samples, performing statistical processing on the values ​​of the fields with non-missing values, and filling the statistical values ​​into the fields with missing values; performing box plot identification on the data samples, determining outliers in the data samples, and performing standard score processing on the outliers; identifying non-numeric fields from the data samples, converting the data of the non-numeric fields to numeric types; and standardizing the numerical features in the data samples.

[0120] In one embodiment, the risk assessment device also includes: a first model updating module, which is used to execute the following steps after the target model is obtained through training: step S1, performing an accuracy test on the output result of the target model through a test set to obtain a first prediction accuracy; step S2, detecting whether the first prediction accuracy is less than a preset accuracy; step S3, when the first prediction accuracy is less than the preset accuracy, adjusting the model parameters of the target model to obtain an adjusted target model, wherein the model parameters include the number of decision trees and / or the depth of the decision trees; step S4, performing an accuracy test on the output result of the adjusted target model through a test set to obtain a second prediction accuracy, and updating the first prediction accuracy to the second prediction accuracy, and repeating steps S2 to S4 until the first prediction accuracy is less than the preset accuracy.

[0121] In one embodiment, the risk assessment device also includes: a second model updating module, which is used to, after training to obtain the target model, perform an accuracy test on the output result of the target model through a test set to obtain a first prediction accuracy; adjust the data features in the test set to obtain an adjusted test set; perform an accuracy test on the output result of the target model through the adjusted test set to obtain a third prediction accuracy; determine the feature importance of the adjusted data features based on the difference between the first prediction accuracy and the third prediction accuracy, wherein the feature importance is used to characterize the degree of influence of the adjusted data features on the prediction accuracy of the target model; when the feature importance is greater than the preset importance, determine that the adjusted data features are key input data of the target model.

[0122] In one embodiment, the business processing module is specifically used to allow the pending business to be processed for the enterprise to be assessed when the risk level is lower than the preset level, and to determine the business processing strategy for the pending business of the enterprise to be assessed according to the level corresponding to the risk level; when the risk level is higher than the preset level, the pending business of the enterprise to be assessed is prohibited.

[0123] In one embodiment, the risk assessment device also includes: a policy adjustment module, which is used to obtain risk level distribution information corresponding to the risk levels of multiple enterprises to be assessed when there are multiple enterprises to be assessed, wherein multiple enterprises to be assessed request to handle pending businesses; and adjust the business handling strategies of the pending businesses at different risk levels according to the risk level distribution information.

[0124] Each module of the risk assessment device provided in the embodiment of the present application can realize Figure 1 The functions of each step of the risk assessment method are provided and can achieve its corresponding technical effects. For the sake of concise description, they will not be repeated here.

[0125] Figure 5 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application is shown.

[0126] The electronic device may include a processor 501 and a memory 502 storing computer program instructions.

[0127] Specifically, the processor 501 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0128] The memory 502 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 502 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 502 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 502 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 502 is a non-volatile solid-state memory.

[0129] The memory may include read-only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical or other physical / tangible memory storage devices. Thus, typically, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of the present application.

[0130] The processor 501 implements any one of the risk assessment methods in the above embodiments by reading and executing computer program instructions stored in the memory 502 .

[0131] In one example, the electronic device may further include a communication interface 503 and a bus 510. Figure 5 As shown, the processor 501, the memory 502, and the communication interface 503 are connected via a bus 510 and communicate with each other.

[0132] The communication interface 503 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0133] Bus 510 includes hardware, software or both, and the parts of electronic equipment are coupled to each other. For example, but not limitation, bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 510 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the application considers any suitable bus or interconnection.

[0134] The risk assessment method in the above embodiment can be implemented by providing a computer storage medium in the embodiment of the present invention. The computer storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any risk assessment method in the above embodiment is implemented.

[0135] An embodiment of the present application also provides a computer program product, including a computer program, which implements any one of the risk assessment methods in the above embodiments when the computer program is processed and executed.

[0136] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0137] The above reference is to the flow chart and / or block diagram of the risk assessment method, device, equipment, storage medium and computer program product according to the embodiment of the present disclosure. Various aspects of the present disclosure are described. It should be understood that each box in the flow chart and / or block diagram and the combination of each box in the flow chart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flow chart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flow chart and the combination of boxes in the block diagram and / or flow chart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0138] The above are only specific implementation methods of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. A risk assessment method, characterized in that: include: Acquire multi-dimensional data corresponding to the enterprise to be evaluated, wherein the multi-dimensional data at least includes historical loan data of the enterprise to be evaluated, operating data of the enterprise to be evaluated, and macroeconomic indicators; Processing the multi-dimensional data through a target model to generate a risk probability of default of the enterprise to be evaluated, wherein the target model is a neural network model obtained by training a training set, and the training set at least includes multi-dimensional data of different enterprises and labels corresponding to the different enterprises, and the labels are used to characterize the risk probability of default of the corresponding enterprise; Determine the risk level corresponding to the enterprise to be assessed under the pending business according to the risk probability and the risk probability ranges of multiple risk levels corresponding to the pending business, wherein the pending business is the business requested by the enterprise to be assessed; Determine whether to handle the pending business for the enterprise to be assessed according to the risk level.

2. The method according to claim 1, characterized in that The target model is trained by the following steps: Acquire a data sample, wherein the data sample at least includes multi-dimensional data corresponding to a plurality of enterprises and risk probabilities corresponding to the plurality of enterprises; Randomly divide the data sample according to a preset ratio to obtain the training set and the test set; Extracting data features corresponding to the multi-dimensional data and labels corresponding to the data features from the training set; Constructing a classifier, and initializing the classifier using a preset number of decision trees and a preset decision tree depth to obtain an initialized classifier; The data features corresponding to the multi-dimensional data and the labels corresponding to the data features are input into the initialized classifier to train the target model.

3. The method according to claim 2, characterized in that After acquiring the data sample, the method further includes: The data sample is preprocessed by any one or more of the following methods: Identify fields with missing values ​​from the data sample, perform statistical processing on the values ​​of fields with non-missing values, and fill the fields with missing values ​​with the statistically obtained values; Performing box plot identification on the data sample, determining outliers in the data sample, and performing standard score processing on the outliers; Identify a non-numeric field from the data sample, and convert the data of the non-numeric field to a numeric type; The numerical features in the data sample are standardized.

4. The method according to claim 2, characterized in that: After the target model is obtained through training, the method further includes: Step S1, performing an accuracy test on the output result of the target model using the test set to obtain a first prediction accuracy; Step S2: Detect whether the first prediction accuracy is less than a preset accuracy. Step S3, when the first prediction accuracy is less than the preset accuracy, adjusting the model parameters of the target model to obtain an adjusted target model, wherein the model parameters include the number of decision trees and / or the depth of the decision tree; Step S4, performing an accuracy test on the output result of the adjusted target model through the test set to obtain a second prediction accuracy, and updating the first prediction accuracy to the second prediction accuracy, and repeating steps S2 to S4 until the first prediction accuracy is less than the preset accuracy.

5. The method according to claim 2, characterized in that: After the target model is obtained through training, the method further includes: Performing an accuracy test on the output result of the target model using the test set to obtain a first prediction accuracy; Adjusting the data features in the test set to obtain an adjusted test set; Performing an accuracy test on the output result of the target model using the adjusted test set to obtain a third prediction accuracy; Determining, according to a difference between the first prediction accuracy and the third prediction accuracy, a feature importance of the adjusted data feature, wherein the feature importance is used to characterize the degree of influence of the adjusted data feature on the prediction accuracy of the target model; When the feature importance is greater than a preset importance, the adjusted data feature is determined to be key input data of the target model.

6. The method according to any one of claims 1 to 5, characterized in that Determining whether to handle the pending business for the enterprise to be assessed according to the risk level includes: In the case where the risk level is lower than the preset level, the pending business is allowed to be processed for the enterprise to be evaluated, and a business processing strategy for the enterprise to be evaluated to process the pending business is determined according to the level corresponding to the risk level; When the risk level is higher than the preset level, it is prohibited to handle the pending business for the enterprise to be assessed.

7. The method according to claim 6, characterized in that In the case where there are multiple enterprises to be evaluated, the method further includes: Obtaining risk level distribution information of risk levels corresponding to a plurality of the to-be-assessed enterprises, wherein a plurality of the to-be-assessed enterprises request to handle the pending business; The business handling strategies for the pending businesses at different risk levels are adjusted according to the risk level distribution information.

8. A risk assessment device, characterized in that: include: A data acquisition module, used to acquire multi-dimensional data corresponding to the enterprise to be evaluated, wherein the multi-dimensional data at least includes historical loan data of the enterprise to be evaluated, operating data of the enterprise to be evaluated, and macroeconomic indicators; A data processing module, used to process the multi-dimensional data through a target model to generate a risk probability of default of the enterprise to be evaluated, wherein the target model is a neural network model obtained by training a training set, and the training set at least includes multi-dimensional data of different enterprises and labels corresponding to the different enterprises, and the labels are used to characterize the risk probability of default of the corresponding enterprise; A level determination module, used to determine the risk level corresponding to the enterprise to be assessed under the pending business according to the risk probability and the risk probability range of multiple risk levels corresponding to the pending business, wherein the pending business is the business requested by the enterprise to be assessed; The business processing module is used to determine whether to process the pending business for the enterprise to be assessed according to the risk level.

9. An electronic device, characterized in that: The electronic device comprises: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the risk assessment method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: Computer program instructions are stored on the computer-readable storage medium, and when the computer program instructions are executed by the processor, the risk assessment method according to any one of claims 1 to 7 is implemented.

11. A computer program product, characterized in that When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device is enabled to perform the risk assessment method as described in any one of claims 1 to 7.

Images