Method and device for adjusting enterprise credit risk score card model, and storage medium

Abstract

The embodiment of the invention provides an adjustment method and device for an enterprise credit risk score card model and a storage medium, and relates to the technical field of model optimization. The method comprises the following steps: determining that a target object score card model meets a hyper-parameter optimization condition, and obtaining corresponding credit score feature data; obtaining a hyper-parameter type of the target object score card model, and determining a hyper-parameter value range; within the hyper-parameter value range of each hyper-parameter type, determining a target hyper-parameter combination enabling the target object score card model to have an optimal prediction performance index based on the credit score feature data by using a preset hyper-parameter optimization strategy; and adjusting the target object score card model based on the target hyper-parameter combination. According to the embodiment of the invention, the optimization condition is set, and the hyper-parameter of the scorecard model is triggered to be optimized and adjusted when the condition is met, so that the timeliness of the scorecard model is improved, and the scorecard model can be ensured to always keep relatively high credit risk prediction capability.

Description

Technical Field

[0001] The present application relates to the technical field of model optimization, and in particular to an adjustment method, device and storage medium for an enterprise credit risk scoring card model. Background Art

[0002] Corporate credit risk models are tools used by financial institutions to assess corporate default risk. These models can help financial institutions make more informed credit decisions and risk management.

[0003] The Carding Model is a common corporate credit risk model. The Carding Model is used to classify customers into different risk levels based on the risk characteristics and behavior patterns of different customers by analyzing historical data.

[0004] Traditional scoring card models can predict future risk probabilities based on specific risk characteristics by analyzing and modeling a large amount of historical data. However, current scoring card models mainly rely on subjective experience to set the model's hyperparameters before machine learning, and once the model is established, it uses fixed hyperparameters for iterative training and prediction tasks, which results in the scoring card model quickly losing its timeliness after modeling and failing to maintain a high credit risk prediction capability for a long time. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a method, device and storage medium for adjusting an enterprise credit risk scoring card model, so as to improve the timeliness of the enterprise credit risk scoring card model, thereby maintaining a high credit risk prediction capability for a long time.

[0006] In a first aspect, an embodiment of the present application provides a method for adjusting an enterprise credit risk scorecard model, comprising:

[0007] Determine whether the target object scorecard model meets the preset hyperparameter optimization conditions, and obtain credit score feature data corresponding to the preset historical period;

[0008] Obtaining a hyperparameter type corresponding to the target object scorecard model, and determining a hyperparameter value range for each hyperparameter type;

[0009] Within the hyperparameter value range of each hyperparameter type, a preset hyperparameter optimization strategy is used to determine, based on the credit score feature data, a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator; wherein the hyperparameter optimization strategy includes at least one of a grid search optimization strategy, a random search optimization strategy, and a Bayesian optimization strategy;

[0010] The target object scorecard model is adjusted based on the target hyperparameter combination.

[0011] In an embodiment of the present application, by setting optimization conditions and triggering optimization adjustment of the hyperparameters of the scoring card model when the conditions are met, the timeliness of the scoring card model is improved to ensure that the scoring card model can always maintain a high credit risk prediction ability.

[0012] In some possible embodiments, determining that the target object scorecard model satisfies a preset hyperparameter optimization condition includes at least one of the following:

[0013] In the case where it is determined that the cumulative unadjusted time of the target object scorecard model exceeds a preset optimization period, determining that the target object scorecard model meets a preset hyperparameter optimization condition;

[0014] In the case where it is determined that the data type of the credit score feature data used by the target object scorecard model has changed, determining that the target object scorecard model satisfies a preset hyperparameter optimization condition;

[0015] In the case where it is determined that the target object scorecard model needs to be retrained, it is determined that the target object scorecard model meets a preset hyperparameter optimization condition.

[0016] In the embodiment of the present application, by setting different hyperparameter optimization conditions to trigger hyperparameter optimization of the model, the timeliness of the scoring card model can be further improved to ensure that the scoring card model can always maintain a high credit risk prediction capability.

[0017] In some possible embodiments, within the hyperparameter value range of each hyperparameter type, using a preset hyperparameter optimization strategy, determining a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator based on the credit score feature data includes:

[0018] Dividing the hyperparameter value range of each hyperparameter type into discrete grid point sets based on a preset step size, and performing traversal and combination based on the grid points in each grid point set to obtain a plurality of candidate hyperparameter combinations corresponding to each hyperparameter type;

[0019] Based on the credit score feature data, a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator is selected from the plurality of candidate hyperparameter combinations.

[0020] In the embodiment of the present application, by using a grid search method to traverse and find the optimal hyperparameter combination, the scoring card model can be adjusted more accurately, further improving the accuracy of the credit score.

[0021] In some possible embodiments, within the hyperparameter value range of each hyperparameter type, using a preset hyperparameter optimization strategy, determining a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator based on the credit score feature data includes:

[0022] Based on the hyperparameter value range of each hyperparameter type, hyperparameter sampling values ​​are randomly selected, and based on the hyperparameter sampling values ​​corresponding to each hyperparameter type, traversal combinations are performed to obtain a number of candidate hyperparameter combinations corresponding to each hyperparameter type;

[0023] Based on the credit score feature data, a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator is selected from the plurality of candidate hyperparameter combinations.

[0024] In the embodiment of the present application, by finding the optimal hyperparameter combination based on random search, the efficiency of hyperparameter optimization can be improved while basically ensuring accuracy.

[0025] In some possible embodiments, within the hyperparameter value range of each hyperparameter type, using a preset hyperparameter optimization strategy, determining a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator based on the credit score feature data includes:

[0026] Within the hyperparameter value range of each hyperparameter type, using a preset grid search optimization strategy, based on the credit score feature data, determine at least two initial hyperparameter combinations that enable the target object scorecard model to have the best prediction performance index;

[0027] Determine an optimized value range corresponding to each of the hyperparameter types based on the at least two initial hyperparameter combinations;

[0028] Within the optimization value range of each of the hyperparameter types, a preset hyperparameter optimization strategy is used to determine a target hyperparameter combination that enables the target object scorecard model to have the best prediction performance index based on the credit score feature data.

[0029] In an embodiment of the present application, a grid search method is used to preliminarily determine and narrow the range of hyperparameter values, and then hyperparameter optimization is further performed based on the optimized hyperparameter value range, thereby improving the efficiency of hyperparameter optimization and the accuracy of optimization.

[0030] In some possible embodiments, the step of selecting a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator from the plurality of candidate hyperparameter combinations based on the credit score feature data includes:

[0031] Based on the credit score feature data, a prediction performance indicator corresponding to each candidate hyperparameter combination is obtained respectively; wherein the prediction performance indicator is obtained by weighted summation based on multiple basic performance indicators, and the basic performance indicators include at least one of the following: accuracy, precision, recall, and a harmonic mean of precision and recall;

[0032] The candidate hyperparameter combination corresponding to the optimal prediction performance indicator is determined as the target hyperparameter combination.

[0033] In the embodiment of the present application, by combining preset weights to obtain prediction performance indicators from multiple dimensions, the flexibility and accuracy of model hyperparameter optimization can be further improved.

[0034] In some possible embodiments, the method for adjusting the enterprise credit risk scorecard model further includes:

[0035] Acquire historical feature data for a preset historical period and corresponding historical credit score results based on the credit score feature type currently bound to the target object scorecard model;

[0036] Calculating the correlation coefficient between each credit score feature type and the target object scorecard model based on the historical feature data and the historical credit score results;

[0037] When the correlation coefficient is lower than a preset threshold, the credit score feature type currently bound to the target object scorecard model is adjusted.

[0038] In the embodiment of the present application, the accuracy and timeliness of the credit score are further improved by obtaining the correlation between each scoring feature and the credit score in real time and adjusting the feature type bound to the model in time according to the correlation.

[0039] In a second aspect, an embodiment of the present application provides a device for adjusting an enterprise credit risk scorecard model, comprising:

[0040] A condition monitoring module is used to determine whether the target object scorecard model meets the preset hyperparameter optimization conditions and obtain credit score feature data corresponding to a preset historical period;

[0041] A range determination module, used to obtain the hyperparameter type corresponding to the target object scorecard model and determine the hyperparameter value range of each hyperparameter type;

[0042] A parameter optimization module, used to determine, within the range of hyperparameter values ​​of each hyperparameter type, a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator based on the credit score feature data by using a preset hyperparameter optimization strategy; wherein the hyperparameter optimization strategy includes at least one of a grid search optimization strategy, a random search optimization strategy, and a Bayesian optimization strategy;

[0043] A model adjustment module is used to adjust the target object scorecard model based on the target hyperparameter combination.

[0044] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor can implement the method described in any embodiment of the first aspect when executing the program.

[0045] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described in any embodiment of the first aspect can be implemented.

[0046] In a fifth aspect, an embodiment of the present application provides a computer program product, wherein the computer program product includes a computer program, wherein when the computer program is executed by a processor, the method described in any embodiment of the first aspect can be implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A flowchart of a method for adjusting an enterprise credit risk scorecard model provided in an embodiment of the present application;

[0049] Figure 2 A schematic diagram of the structure of the confusion matrix provided in the embodiment of the present application;

[0050] Figure 3 A schematic diagram of the structure of an adjustment device for an enterprise credit risk scorecard model provided in an embodiment of the present application;

[0051] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0053] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0054] It should be noted that traditional scoring card models mainly rely on human subjective experience to set the model's hyperparameters before machine learning, and once the model is established, fixed hyperparameters are used for iterative training and prediction tasks. Therefore, in the rapidly changing environment of the financial market, the scoring card model may quickly lose its timeliness after modeling and cannot maintain a high level of credit risk prediction capabilities for a long time.

[0055] To this end, an embodiment of the present application provides a method for adjusting an enterprise credit risk scoring card model, which ensures that the scoring card model always maintains a high degree of timeliness by setting optimization conditions and optimizing the hyperparameters of the scoring card model when the conditions are met, thereby improving the accuracy of credit risk prediction.

[0056] like Figure 1 As shown, the embodiment of the present application provides a method for adjusting an enterprise credit risk scorecard model, which may include the steps of:

[0057] S1. Determine whether the target object scorecard model satisfies preset hyperparameter optimization conditions, and obtain credit score feature data corresponding to a preset historical period.

[0058] Specifically, the scoring card model of the target object is monitored in real time to see if it meets the preset hyperparameter optimization conditions. When it is judged that the conditions are met, the hyperparameter optimization of the scoring card model of the target object is started. The target object is one of multiple monitored objects, and the monitored objects refer to all enterprises that need to undergo credit risk assessment. Different monitored objects can correspond to different scoring card models, which can be classified and bound according to the industry attributes or types of the enterprises. In addition, different scoring card models can correspond to different hyperparameter optimization conditions. The hyperparameter optimization conditions can be pre-set according to the needs, for example, set to perform regular optimization on a periodic basis.

[0059] When it is determined that the conditions are met, that is, when the hyperparameter adjustment of the target object scorecard model is required, the credit score feature data corresponding to the target object in the preset historical period is first obtained based on the current moment. Among them, the feature type of the credit score feature data is predetermined according to the actual situation, for example, it may include financial indicators, operating conditions, industry characteristics and other features.

[0060] S2. Obtain the hyperparameter type corresponding to the target object scorecard model, and determine the hyperparameter value range of each hyperparameter type.

[0061] Specifically, the hyperparameter types that need to be optimized corresponding to the target object scorecard model are determined, and then the hyperparameter value ranges of each hyperparameter type are determined respectively.

[0062] It is understandable that the target object scorecard model can adopt a variety of machine learning models such as logistic regression, decision tree, random forest, support vector machine, etc. Different models have different hyperparameter types, and for different target object scorecard models, it is possible to choose which hyperparameter types to optimize. For example, the number of trees and the maximum depth in the random forest can be set as the hyperparameter types that need to be optimized; for another example, the kernel function and penalty parameter in the support vector machine can be set as the hyperparameter types that need to be optimized.

[0063] It should be noted that for different hyperparameter types, the corresponding hyperparameter value range can be predetermined. Exemplarily, for a logistic regression model, its regularization parameter is used to control the complexity of the model to prevent overfitting, and its value range is usually between 0.01 and 100; for a decision tree, its maximum depth is used to limit the growth depth of the tree to prevent overfitting, and its value range is usually between 2 and 20; for a random forest, the number of trees usually ranges from 10 to 1000; for a support vector machine, different kernel functions have different parameters that need to be adjusted: for example, for a radial basis function (RBF) kernel, the gamma parameter is used to control the width of the kernel, and its value range is usually between 0.001 and 10. In practical applications, the value range corresponding to each hyperparameter type can be determined by referring to relevant literature, using empirical methods, or conducting preliminary experiments.

[0064] S3. Within the range of hyperparameter values ​​for each hyperparameter type, a preset hyperparameter optimization strategy is used to determine a target hyperparameter combination that enables the target object scorecard model to have the best prediction performance index based on the credit score feature data; wherein the hyperparameter optimization strategy includes at least one of a grid search optimization strategy, a random search optimization strategy, and a Bayesian optimization strategy.

[0065] Specifically, within the corresponding hyperparameter value range, the preset hyperparameter optimization strategy (such as grid search, random search or Bayesian optimization, etc.) is used, and the credit score feature data is used as the training set and verification set to train and predict the target object score card model, and the optimal target hyperparameter combination is determined based on the prediction performance index. Among them, the prediction performance index can be the weighted value of multiple indicators, such as accuracy, precision, recall rate, etc. The contribution (weight) of each indicator to the model performance is set according to the needs, and then these indicators are weighted summed to obtain the final prediction performance index.

[0066] S4. Adjust the target object scorecard model based on the target hyperparameter combination.

[0067] Finally, the target hyperparameter combination of the target object scoring card model with the optimal prediction performance indicator is set as the target value to adjust the hyperparameters of the original target object scoring card model (for example, directly replace the target value).

[0068] Based on this, by setting optimization conditions and optimizing and adjusting the hyperparameters of the scoring card model in a timely manner when the conditions are met, the timeliness of the scoring card model is effectively improved to ensure that the scoring card model can always maintain a high credit risk prediction ability.

[0069] It should be noted that for different types of enterprises, appropriate machine learning models can be selected as enterprise credit risk scorecard models according to their respective characteristics. Specifically, model selection can be based on data characteristics: 1. Based on the size of the data; if the data size is small, logistic regression and support vector machines may be more suitable; if the data size is large, decision trees and random forests may perform better. 2. Based on data feature dimensions; if the feature dimension is high, random forests and support vector machines may be better at processing high-dimensional data; if the feature dimension is low, logistic regression and decision trees may be simpler and more effective. 3. Based on data distribution: If the data distribution is more complex and there are nonlinear relationships, decision trees, random forests and support vector machines may be more suitable; if the data distribution is relatively simple, logistic regression may be sufficient. 4. Data quality: If there are more missing values ​​and outliers in the data, decision trees and random forests may be more robust.

[0070] In addition, you can also choose an appropriate model based on the business needs of the enterprise: 1. Based on the requirements of interpretability: If the results of the model need to be explained, logistic regression and decision trees may be more suitable because their model structures are relatively simple and easy to understand. 2. Based on the requirements of accuracy: If the accuracy requirements of the model are high, you can try multiple models and select the model with the best performance through methods such as cross-validation. 3. Based on the requirements of real-time: If you need to make a credit risk assessment quickly, logistic regression and decision trees may be faster in calculation and more suitable for real-time applications. 4. Based on scalability requirements: If the model needs to be able to adapt to changing data and business needs, random forests and support vector machines may be more flexible.

[0071] In some possible embodiments, in step S1, determining whether the target object scorecard model satisfies a preset hyperparameter optimization condition includes at least one of the following:

[0072] S101: when it is determined that the cumulative unadjusted time of the target object scorecard model exceeds a preset optimization period, determining that the target object scorecard model meets a preset hyperparameter optimization condition;

[0073] S102: when it is determined that the data type of the credit score feature data used by the target object scorecard model has changed, determining that the target object scorecard model satisfies a preset hyperparameter optimization condition;

[0074] S103: When it is determined that the target object scoring card model needs to be retrained, determine whether the target object scoring card model meets a preset hyperparameter optimization condition.

[0075] It should be noted that, in step S101, the hyperparameters of the target object scoring card model can be optimized and updated regularly according to a preset optimization cycle; specifically, the cumulative duration of the target object scoring card model without hyperparameter adjustment can be recorded in real time, and when the duration exceeds the preset optimization cycle, it is determined that the preset hyperparameter optimization conditions are met.

[0076] In step S102, the data type (feature type) of the credit scoring feature data used by the target object scoring card model may need to be adjusted accordingly as the market environment changes. In order to ensure that the model can adapt to the latest market changes and corporate behavior changes, after adjusting the feature type used by the target object scoring card model, it is necessary to first optimize and update the hyperparameters of the target object scoring card model before performing credit risk assessment.

[0077] In step S103, in order to ensure that the model can adapt to the latest data changes, it is usually necessary to retrain the model regularly or based on specific event triggers, so that the retrained model can maintain a high timeliness to improve the model's prediction ability. When it is determined that the target object scorecard model needs to be retrained, the hyperparameters of the model can be optimized and adjusted first, and then the model can be retrained after the hyperparameters are optimized and adjusted.

[0078] Based on this, by setting different hyperparameter optimization conditions, the hyperparameter optimization of the model can be triggered in time under different circumstances, thereby further improving the timeliness of the scoring card model to ensure that the scoring card model can always maintain a high credit risk prediction ability.

[0079] In some possible embodiments, step S3, within the hyperparameter value range of each hyperparameter type, using a preset hyperparameter optimization strategy, and based on the credit score feature data, determining a target hyperparameter combination that enables the target object scorecard model to have the optimal prediction performance index, may include:

[0080] S301, dividing the hyperparameter value range of each hyperparameter type into discrete grid point sets based on a preset step size, and performing traversal and combination based on the grid points in each grid point set to obtain a number of candidate hyperparameter combinations corresponding to each hyperparameter type;

[0081] S302: Based on the credit score feature data, a target hyperparameter combination is selected from a plurality of candidate hyperparameter combinations so that the target object scorecard model has the best prediction performance index.

[0082] It should be noted that steps S301 to S302 are the process of optimizing hyperparameters using a grid search optimization strategy. Grid search is a systematic hyperparameter tuning method that exhaustively searches a predefined hyperparameter space (hyperparameter value range) to ultimately find the best hyperparameter combination. Specifically, the grid search lists all possible hyperparameter combinations, then performs model training and evaluation on each combination, and finally selects the combination with the best performance on the validation set. For each combination, a cross-validation method or other method can be used to evaluate model performance, such as the K-fold cross-validation method: the data set is divided into K subsets, each subset is used as a validation set in turn, and the other subsets are used as training sets, thereby ensuring the generalization ability of the model.

[0083] Among them, when the value range of each hyperparameter is divided into discrete grid points, the step size of the division can be set according to the needs. Generally, the larger the step size, the higher the efficiency of the grid search optimization, and the smaller the step size, the higher the accuracy of the grid search optimization. It can be understood that the advantage of the grid search optimization method is that it can traverse all possible parameter combinations to ensure that the global optimal solution is found, but its disadvantage is that the computational cost is high. When the number of hyperparameters is large or the value range is large, the optimization calculation time is long.

[0084] In some possible embodiments, step S3, within the hyperparameter value range of each hyperparameter type, using a preset hyperparameter optimization strategy, and based on the credit score feature data, determining a target hyperparameter combination that enables the target object scorecard model to have the optimal prediction performance index, may include:

[0085] S311, randomly selecting hyperparameter sampling values ​​based on the hyperparameter value range of each hyperparameter type, and performing traversal combinations based on the hyperparameter sampling values ​​corresponding to each hyperparameter type to obtain a number of candidate hyperparameter combinations corresponding to each hyperparameter type;

[0086] S312. Based on the credit score feature data, a target hyperparameter combination is selected from a number of candidate hyperparameter combinations so that the target object scorecard model has the best prediction performance index.

[0087] It should be noted that steps S311 to S312 are the process of using random search optimization strategy to optimize hyperparameters. Random search is a hyperparameter tuning method that randomly samples multiple hyperparameter combinations in a predefined hyperparameter space, trains and evaluates each combination, and finds the best performing hyperparameter combination. Unlike grid search, random search does not exhaust all possible combinations, but randomly selects a portion of combinations for evaluation.

[0088] Specifically, the number of random sampling is first determined, and then random sampling is performed based on the range of hyperparameter values. For each sampling, a set of candidate hyperparameter combinations is randomly generated; then the model is trained and evaluated for each candidate hyperparameter combination, and finally the hyperparameter combination with the best performance is found.

[0089] It is understandable that random search only evaluates some hyperparameter combinations within the value range, and the computational cost is much lower than that of grid search. At the same time, due to its high search efficiency, random search can usually find near-optimal hyperparameter combinations faster in high-dimensional hyperparameter space. However, the random search method cannot guarantee the global optimal solution and may miss some hyperparameter combinations with good performance.

[0090] In some possible embodiments, step S3, within the hyperparameter value range of each hyperparameter type, using a preset hyperparameter optimization strategy, and based on the credit score feature data, determining a target hyperparameter combination that enables the target object scorecard model to have the optimal prediction performance index, may include:

[0091] S321. Within the range of hyperparameter values ​​of each hyperparameter type, using a preset grid search optimization strategy, determine at least two initial hyperparameter combinations that enable the target object scorecard model to have the best prediction performance index based on the credit score feature data;

[0092] S322, determining an optimized value range corresponding to each hyperparameter type based on at least two initial hyperparameter combinations;

[0093] S323. Within the optimization value range of each hyperparameter type, using a preset hyperparameter optimization strategy, determine a target hyperparameter combination that enables the target object scorecard model to have the best prediction performance index based on the credit score feature data.

[0094] It should be noted that a grid search optimization strategy can be used based on a larger step size to preliminarily determine a better hyperparameter value range (optimized value range). This process is equivalent to rapidly narrowing the hyperparameter value range; then, based on the optimized value range, the above optimization strategy (grid search, random search or Bayesian optimization, etc.) is used to optimize the hyperparameters at a finer granularity. In this process, an iterative optimization method can be used to continuously narrow the search range of the hyperparameters until a sufficiently fine granularity is used to optimize and obtain a nearly optimal hyperparameter combination.

[0095] It should be noted that the principle of Bayesian Optimization is to construct a proxy model to approximate the objective function (model performance), and use the acquisition function to determine the next hyperparameter combination to be evaluated. As the evaluation proceeds, the proxy model and acquisition function are continuously updated to guide the search to more promising areas. Hyperparameter optimization through Bayesian optimization can efficiently search high-dimensional hyperparameter space, and usually find better solutions faster than grid search and random search.

[0096] Based on this, by preliminarily determining and narrowing the range of hyperparameter values, and then further refining the hyperparameter optimization based on the optimized hyperparameter value range, the efficiency of hyperparameter optimization is improved while the accuracy of optimization is improved.

[0097] In some possible embodiments, step S302 or S312, screening out a target hyperparameter combination that enables the target object scorecard model to have the best prediction performance index from a number of candidate hyperparameter combinations based on the credit score feature data, may include:

[0098] Based on the credit score feature data, a prediction performance indicator corresponding to each candidate hyperparameter combination is obtained respectively; wherein the prediction performance indicator is obtained by weighted summation based on multiple basic performance indicators, and the basic performance indicators include at least one of the following: accuracy, precision, recall, and a harmonic mean of precision and recall;

[0099] The candidate hyperparameter combination corresponding to the optimal prediction performance indicator is determined as the target hyperparameter combination.

[0100] It should be noted that in each model prediction process, the relevant performance indicators of the model can be calculated based on the actual value and predicted value of the credit score feature data, among which the basic performance indicators can include accuracy, precision, recall, and the harmonic mean of precision and recall.

[0101] When analyzing the classification of prediction results, the confusion matrix can be combined to identify the advantages and disadvantages of the model. The confusion matrix clearly shows the classification of the model prediction results and the actual results in a tabular form. For the corporate credit risk scorecard model, companies are usually divided into high risk (positive class) and low risk (negative class). The four cells in the confusion matrix represent true positive (TP), false positive (FP), true negative (TN) and false negative (FN).

[0102] like Figure 2 As shown, TP represents the number of samples that are actually high risk and correctly predicted as high risk by the model; FP represents the number of samples that are actually low risk but incorrectly predicted as high risk by the model; TN represents the number of samples that are actually low risk and correctly predicted as low risk by the model; FN represents the number of samples that are actually high risk but incorrectly predicted as low risk by the model.

[0103] Accuracy is used to measure the proportion of correct predictions of the model as a whole. For the credit risk scorecard model, a high accuracy rate means that the credit risk status of the enterprise can be accurately judged. The accuracy calculation formula is: Understandably, the basic performance indicator of accuracy intuitively reflects the overall predictive ability of the model, but in the case of unbalanced data sets or when there is a high focus on specific categories, this indicator may not fully reflect the performance of the model; for example, if the proportion of negative samples (high-risk enterprises) is very low, even if the model always predicts positive samples (low-risk enterprises), the accuracy may be high, but in fact the model does not identify high-risk enterprises well.

[0104] Precision refers to the proportion of positive samples among the results predicted as positive samples. The formula for calculating precision is: In the credit risk scorecard model, a high precision rate means that the model is more accurate in predicting high-risk enterprises, reducing the misjudgment of low-risk enterprises as high-risk enterprises, which is of great significance for prudent risk control and avoiding unnecessary risk warnings. The precision rate indicator mainly focuses on the model's prediction accuracy for positive samples, but may ignore the prediction of negative samples.

[0105] Recall refers to the proportion of samples that are correctly predicted as positive samples among the results that are actually positive samples. The calculation formula is: In the credit risk scorecard, a high recall rate means that the model can identify as many high-risk companies as possible. The recall rate focuses on the model's coverage of high-risk companies. A high recall rate can ensure that as many truly high-risk companies as possible are identified, which is crucial for timely detection of potential risks and taking corresponding measures.

[0106] The harmonic mean of precision and recall is calculated as: This indicator takes into account the balance between precision and recall, and is an important evaluation indicator when both accuracy and coverage need to be taken into account. In the credit risk scorecard model, a high F1 value means that the model has achieved a good balance between accuracy and coverage.

[0107] It should be noted that the confusion matrix can be used to identify the advantages and disadvantages of the model during the model adjustment process. If the data set is unbalanced, that is, the proportion of high-risk enterprises and low-risk enterprises is quite different, the confusion matrix can help find the performance differences of the model in different categories. For example, if the number of high-risk enterprises is small, the model may tend to predict more enterprises as low-risk, resulting in a lower recall rate. By analyzing the confusion matrix, the model can be adjusted in a targeted manner, such as using oversampling, undersampling or synthetic minority class oversampling techniques to process unbalanced data and improve the model's recognition ability for minority classes.

[0108] In addition, the confusion matrix can reveal the types of errors in the model and help identify the main problems with the model. If the FP is high, it means that the model has more false positives, which may cause unnecessary costs and interference to the company; if the FN is high, it means that the model has missed more high-risk companies, which may cause potential risks to not be discovered in time. Based on the analysis results of the error type, the model's hyperparameters can be adjusted, feature selection can be improved, or different machine learning algorithms can be used to reduce specific types of errors.

[0109] In addition, the results of the confusion matrix can also provide a certain degree of explainability for the model's decision-making. By analyzing the predictions between different categories, we can understand how the model is classified and which features play a key role in the classification results. This helps business personnel understand the model's decision-making process, enhance trust in the model, and provide direction for further model optimization.

[0110] In some possible embodiments, the method for adjusting the enterprise credit risk scorecard model may further include:

[0111] Based on the credit score feature type currently bound to the target object scorecard model, historical feature data for a preset historical period and corresponding historical credit score results are obtained;

[0112] Calculate the correlation coefficient between each credit score feature type and the target object scorecard model based on historical feature data and historical credit score results;

[0113] When the correlation coefficient is lower than the preset threshold, the credit score feature type currently bound to the target object scorecard model is adjusted.

[0114] It should be noted that in the process of using the enterprise credit risk scorecard model to conduct credit risk assessment, the correlation of the credit scoring feature types used in the model can be monitored in real time, and timely adjustments can be made when the correlation becomes low.

[0115] Specifically, first, the historical feature data of the credit score feature type currently bound to the model in a preset historical period (based on the current moment) can be obtained based on a preset cycle, and the historical credit score results of the same historical period can be obtained. Then, based on these data, the correlation between each credit score feature type and the credit score is calculated, such as the Pearson correlation coefficient, the Spearman correlation coefficient, etc. When the correlation coefficient is lower than the preset threshold, it means that the correlation between the credit score feature type and the credit score is low, and the credit score feature type bound to the model needs to be adjusted to re-determine the credit score feature type with a higher correlation to the model's credit score.

[0116] It should be noted that in order to improve the prediction accuracy, stability and interpretability of the model, feature selection is one of the key steps. The feature types with high correlation with the credit scoring model can be selected based on the following methods:

[0117] 1. Through in-depth understanding of the business processes and key factors of corporate credit risk assessment, determine the potential characteristics that may affect corporate credit risk, such as financial indicators, operating conditions, industry characteristics, etc. You can also consider the characteristics and risk factors of different industries and select characteristics related to the industry to which the target company belongs.

[0118] 2. Based on business understanding, a number of characteristics that may have an impact on the company's credit risk are initially screened out. These characteristics may include financial indicators such as the company's asset size, debt level, profitability, debt repayment ability, and operational efficiency, as well as non-financial indicators such as the company's industry position, market share, and management level.

[0119] 3. Based on the univariate analysis method, calculate the correlation coefficient between each initially screened feature and the target variable (credit risk score), such as the Pearson correlation coefficient, Spearman correlation coefficient, etc., and select the features with a higher correlation with the target variable. Then, perform an independence test on each feature, such as the chi-square test, variance analysis, etc., and select features that are independent of the target variable.

[0120] 4. Based on multivariate analysis methods, such as stepwise regression, ridge regression, lasso regression and other methods, multivariate analysis can be performed. Based on these methods, features that have a significant impact on the target variable can be automatically selected and redundant features can be removed. In addition, dimensionality reduction methods such as principal component analysis (PCA) and factor analysis can be used to transform multiple related features into a few unrelated comprehensive features. These comprehensive features can retain most of the information of the original features while reducing the number of features.

[0121] 5. Based on the filtering method, use indicators such as information gain, mutual information, and Gini index to evaluate the ability of each feature to distinguish the target variable and select features with stronger distinguishing ability. You can also use algorithms such as ReliefF and mRMR for feature selection. These algorithms can consider the correlation between features and select features that are highly correlated with the target variable and independent of each other.

[0122] 6. Based on the wrapper method, greedy algorithms, genetic algorithms and other optimization algorithms are used in combination with specific machine learning models (such as decision trees, support vector machines, etc.) for feature selection. These methods continuously add or delete features to find the optimal feature subset; then use the cross-validation method to evaluate the performance of the model under different feature subsets and select the feature subset with the best performance.

[0123] 7. Based on the embedded method, feature selection is automatically performed during the training process of the machine learning model. For example, in the decision tree model, features with higher importance are selected by calculating the importance score of the features. Regularization techniques such as L1 regularization and L2 regularization can also be used to constrain the parameters of the model to achieve feature selection. L1 regularization can make the coefficients of some features become zero, thereby achieving feature selection; L2 regularization can make the coefficients of the features smaller, thereby reducing the importance of the features.

[0124] It should be noted that in the process of selecting key features, you can use indicators such as accuracy, recall, and F1 value to evaluate the performance of the model after feature selection, and then select the feature subset with the best performance as the key feature. Considering the stability and interpretability of the model, when selecting feature subsets, you should not only pay attention to the performance of the model, but also consider the stability and interpretability of the features. A stable feature subset can improve the generalization ability of the model, and a feature subset with strong interpretability can help business personnel better understand the decision-making process of the model.

[0125] In addition, the cross-validation method can also be used to evaluate the stability and reliability of feature selection. Specifically, the data set is divided into multiple subsets, feature selection and model training are performed on different subsets, and then the performance of the model is evaluated. If the feature selection results and model performance on different subsets are relatively stable, it means that the feature selection method is reliable.

[0126] It should be noted that in the process of collecting data, the embodiments of the present application may pre-process the data, such as cleaning the data, processing missing values ​​and outliers, etc., to ensure the quality of the data.

[0127] For example, the deletion method can be used to deal with missing values: if the proportion of missing values ​​is relatively low, you can consider directly deleting the records containing missing values. However, this method may result in a reduction in the amount of data and affect the accuracy of the model. For a specific variable, if the proportion of missing values ​​is high and the importance of the variable to credit risk assessment is low, you can consider deleting the variable.

[0128] For example, the missing values ​​can also be filled by the filling method: for numerical variables, the mean or median of the variable can be used to fill the missing values. This method is suitable for cases where the data distribution is relatively symmetrical. For categorical variables, the mode of the variable can be used to fill the missing values. In addition, linear interpolation, polynomial interpolation and other methods can also be used to fill the missing values. This method is suitable for cases where the data has certain trends or regularities.

[0129] Exemplarily, the following methods can be used to handle outliers: 1. Statistical methods; use statistical methods such as box plots and the 3σ principle to identify outliers. Box plots can intuitively display the distribution of data, and outliers are usually defined as data points outside the upper and lower boundaries of the box plot. The 3σ principle is based on the assumption of normal distribution, and data points that are more than 3 times the standard deviation from the mean are considered outliers. 2. Business logic judgment method, based on business knowledge and experience, judge whether the outliers in the data are reasonable. For example, if the debt-to-asset ratio of a certain enterprise exceeds 100%, this may be an outlier, but it may also be caused by the special operating conditions of the enterprise. In this case, it is necessary to further analyze the specific situation of the enterprise to determine whether it should be treated as an outlier. 3. Deleting outliers: If the number of outliers is small, you can consider deleting these outliers directly. However, this method may result in a reduction in the amount of data and affect the accuracy of the model. 4. Replace outliers: Statistics such as mean, median, and mode can be used to replace outliers. This method is suitable for situations where the number of outliers is small and the impact on data distribution is small. 5. For continuous variables, they can be binned and outliers can be assigned to specific bins. This method can reduce the impact of outliers on the model while retaining the distribution information of the data.

[0130] It should be noted that in order to eliminate the dimensional differences between different variables, the data needs to be standardized. Common standardization methods are: 1. Minimum-maximum standardization: Map the data to the interval [0,1], and the formula is Where x is the original data, x min is the minimum value of the data, x max 2. Z-score standardization: convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The formula is: Where x is the original data, μ is the mean of the data, and σ is the standard deviation of the data.

[0131] After completing data preprocessing, the cleaned data needs to be verified and quality assessed, such as the following methods: 1. Data visualization: Use charts (such as histograms, scatter plots, etc.) to intuitively display the distribution of data and check whether the data is abnormal or unreasonable; 2. Calculate statistics such as the mean, median, and standard deviation of the data to check whether the central trend and dispersion of the data are reasonable; 3. Use the cleaned data to establish a credit risk scorecard model, and verify and evaluate the model. If the model performs poorly, it may be necessary to re-examine the data cleaning process, identify the problem and make improvements.

[0132] It should be noted that in the process of optimizing and adjusting the hyperparameters of the scoring card model, various performance indicators such as accuracy, recall, F1-score, AUC, etc. can be used to evaluate the model performance.

[0133] It is understandable that model evaluation can provide a basis for hyperparameter optimization: 1. By evaluating the model under different hyperparameter combinations, we can understand the influence trend of each hyperparameter on model performance. For example, if it is found in the evaluation that as the value of a certain hyperparameter increases, the accuracy of the model continues to increase, but the recall rate gradually decreases, this provides a clear direction for hyperparameter optimization. At this time, according to specific business needs, the importance of accuracy and recall can be weighed to adjust the value range of the hyperparameter to achieve a better performance balance. 2. Model evaluation indicators can help determine whether the model has overfitting or underfitting problems. If the performance is good on the training set, but the performance on the test set decreases significantly, it means that the model may be overfitting. At this time, the regularization degree of the model can be increased by adjusting the hyperparameters, such as increasing the regularization parameter C in logistic regression or reducing the maximum depth in decision trees and random forests, to prevent overfitting. On the contrary, if the model performs poorly on both the training set and the test set, there may be underfitting. You can consider adjusting the hyperparameters to increase the complexity of the model, such as increasing the depth of the tree, increasing the number of trees in the random forest, etc.

[0134] In addition, model evaluation not only focuses on traditional performance indicators, but also needs to be combined with actual business needs. For example, in the enterprise credit risk scorecard model, more attention may be paid to the accurate identification of high-risk enterprises, that is, the recall rate. In this case, the weights of the evaluation indicators can be adjusted according to business needs, and more attention can be paid to improving the recall rate during the hyperparameter optimization process. At the same time, you can also consider introducing business-specific constraints, such as requiring the complexity of the model not to be too high to ensure the interpretability and stability of the model. The adjustment of these business needs will directly affect the optimization direction and strategy of the hyperparameters.

[0135] Please refer to Figure 3 , Figure 3 The following is a block diagram showing the composition of the adjustment device for the enterprise credit risk score card model provided by some embodiments of the present application. It should be understood that the adjustment device for the enterprise credit risk score card model is similar to the above-mentioned Figure 1 Corresponding to the method embodiment, each step involved in the above method embodiment can be executed. The specific functions of the adjustment device of the enterprise credit risk scorecard model can be found in the description above. To avoid repetition, the detailed description is appropriately omitted here.

[0136] Figure 3The adjustment device of the enterprise credit risk scorecard model includes at least one software function module that can be stored in a memory in the form of software or firmware or fixed in the adjustment device of the enterprise credit risk scorecard model, and the adjustment device of the enterprise credit risk scorecard model includes:

[0137] The condition monitoring module 310 is used to determine whether the target object scorecard model meets the preset hyperparameter optimization conditions and obtain credit score feature data corresponding to the preset historical period;

[0138] A range determination module 320 is used to obtain the hyperparameter type corresponding to the target object scorecard model and determine the hyperparameter value range of each hyperparameter type;

[0139] The parameter optimization module 330 is used to determine, within the range of hyperparameter values ​​of each hyperparameter type, a target hyperparameter combination that enables the target object scorecard model to have the best prediction performance index based on the credit score feature data by using a preset hyperparameter optimization strategy; wherein the hyperparameter optimization strategy includes at least one of a grid search optimization strategy, a random search optimization strategy, and a Bayesian optimization strategy;

[0140] The model adjustment module 340 is used to adjust the target object scorecard model based on the target hyperparameter combination.

[0141] It can be understood that the above-mentioned device item embodiment corresponds to the method item embodiment of the present invention. The adjustment device of the enterprise credit risk scoring card model provided by the embodiment of the present invention can implement the adjustment method of the enterprise credit risk scoring card model provided by any method item embodiment of the present invention.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method, and will not be described in detail here.

[0143] like Figure 4 As shown, some embodiments of the present application provide an electronic device 400, which includes: a memory 410, a processor 420, and a computer program stored in the memory 410 and executable on the processor 420, wherein the processor 420 reads the program from the memory 410 through a bus 430 and executes the program to implement a method of any embodiment of the method for adjusting the enterprise credit risk scorecard model as described above.

[0144] Processor 420 can process digital signals and can include various computing structures, such as complex instruction set computer structure, reduced instruction set computer structure, or a structure that implements a combination of multiple instruction sets. In some examples, processor 420 can be a microprocessor.

[0145] The memory 410 may be used to store instructions executed by the processor 420 or data related to the execution of instructions. These instructions and / or data may include codes for implementing some or all functions of one or more modules described in the embodiments of the present application. The processor 420 of the disclosed embodiment may be used to execute instructions in the memory 410 to implement the method shown above. The memory 410 includes a dynamic random access memory, a static random access memory, a flash memory, an optical memory, or other memory known to those skilled in the art.

[0146] Some embodiments of the present application further provide a computer-readable storage medium having a computer program stored thereon. The computer program is executed by a processor to execute the method described in the method embodiment.

[0147] Some embodiments of the present application further provide a computer program product, which, when executed on a computer, enables the computer to execute the method described in the method embodiment.

[0148] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0149] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0150] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0151] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0152] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0153] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0154] 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 sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

Claims

1. A method for adjusting an enterprise credit risk scorecard model, characterized in that: include: Determine whether the target object scorecard model meets the preset hyperparameter optimization conditions, and obtain credit score feature data corresponding to the preset historical period; Obtaining a hyperparameter type corresponding to the target object scorecard model, and determining a hyperparameter value range for each hyperparameter type; Within the hyperparameter value range of each hyperparameter type, a preset hyperparameter optimization strategy is used to determine, based on the credit score feature data, a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator; wherein the hyperparameter optimization strategy includes at least one of a grid search optimization strategy, a random search optimization strategy, and a Bayesian optimization strategy; The target object scorecard model is adjusted based on the target hyperparameter combination.

2. The method for adjusting the enterprise credit risk scorecard model according to claim 1, characterized in that: Determining that the target object scorecard model satisfies a preset hyperparameter optimization condition includes at least one of the following: In the case where it is determined that the cumulative unadjusted time of the target object scorecard model exceeds a preset optimization period, determining that the target object scorecard model meets a preset hyperparameter optimization condition; In the case where it is determined that the data type of the credit score feature data used by the target object scorecard model has changed, determining that the target object scorecard model satisfies a preset hyperparameter optimization condition; In the case where it is determined that the target object scorecard model needs to be retrained, it is determined that the target object scorecard model meets a preset hyperparameter optimization condition.

3. The method for adjusting the enterprise credit risk scorecard model according to claim 1, characterized in that: The method of determining, within the hyperparameter value range of each hyperparameter type, a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator based on the credit score feature data by using a preset hyperparameter optimization strategy includes: Dividing the hyperparameter value range of each hyperparameter type into discrete grid point sets based on a preset step size, and performing traversal and combination based on the grid points in each grid point set to obtain a plurality of candidate hyperparameter combinations corresponding to each hyperparameter type; Based on the credit score feature data, a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator is selected from the plurality of candidate hyperparameter combinations.

4. The method for adjusting the enterprise credit risk scorecard model according to claim 1, characterized in that: The method of determining, within the hyperparameter value range of each hyperparameter type, a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator based on the credit score feature data by using a preset hyperparameter optimization strategy includes: Based on the hyperparameter value range of each hyperparameter type, hyperparameter sampling values ​​are randomly selected, and based on the hyperparameter sampling values ​​corresponding to each hyperparameter type, traversal combinations are performed to obtain a number of candidate hyperparameter combinations corresponding to each hyperparameter type; Based on the credit score feature data, a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator is selected from the plurality of candidate hyperparameter combinations.

5. The method for adjusting the enterprise credit risk scorecard model according to claim 1, characterized in that: The method of determining, within the hyperparameter value range of each hyperparameter type, a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator based on the credit score feature data by using a preset hyperparameter optimization strategy includes: Within the hyperparameter value range of each hyperparameter type, using a preset grid search optimization strategy, based on the credit score feature data, determine at least two initial hyperparameter combinations that enable the target object scorecard model to have the best prediction performance index; Determine an optimized value range corresponding to each of the hyperparameter types based on the at least two initial hyperparameter combinations; Within the optimization value range of each of the hyperparameter types, a preset hyperparameter optimization strategy is used to determine a target hyperparameter combination that enables the target object scorecard model to have the best prediction performance index based on the credit score feature data.

6. The method for adjusting the enterprise credit risk scorecard model according to claim 3 or 4, characterized in that: The step of selecting a target hyperparameter combination that enables the target object scorecard model to have an optimal prediction performance indicator from the plurality of candidate hyperparameter combinations based on the credit score feature data includes: Based on the credit score feature data, a prediction performance indicator corresponding to each candidate hyperparameter combination is obtained respectively; wherein the prediction performance indicator is obtained by weighted summation based on multiple basic performance indicators, and the basic performance indicators include at least one of the following: accuracy, precision, recall, and a harmonic mean of precision and recall; The candidate hyperparameter combination corresponding to the optimal prediction performance indicator is determined as the target hyperparameter combination.

7. The method for adjusting the enterprise credit risk scorecard model according to claim 1, characterized in that: Also includes: Acquire historical feature data of a preset historical period and corresponding historical credit score results based on the credit score feature type currently bound to the target object scorecard model; Calculating the correlation coefficient between each credit score feature type and the target object scorecard model based on the historical feature data and the historical credit score results; When the correlation coefficient is lower than a preset threshold, the credit score feature type currently bound to the target object scorecard model is adjusted.

8. An electronic device, characterized in that: It comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for adjusting the enterprise credit risk scorecard model as described in any one of claims 1 to 7 can be implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for adjusting the enterprise credit risk scorecard model according to any one of claims 1 to 7 is executed.

10. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a processor, the method for adjusting the enterprise credit risk scorecard model according to any one of claims 1 to 7 is implemented.

Images