Credit risk control modeling method and system

Abstract

The invention provides a credit risk control modeling method and system, and the method comprises the steps: 1, obtaining a business scene demand of credit risk control modeling, and determining a risk control modeling data set, an initial model, an evaluation index and convergence logic according to the business scene demand; 2, training an initial model according to the risk control modeling data set to obtain an intermediate model, and performing quantitative calculation on evaluation indexes; 3, when the quantitative calculation result of the evaluation index does not meet the convergence logic, performing iterative calculation on the intermediate model according to a preset method, and when the iterative calculation reaches the convergence logic, stopping the iterative calculation; 4, training the target intermediate model obtained by each iteration training according to a preset ensemble learning algorithm to obtain a final credit risk control model; the accuracy and stability of credit risk assessment are effectively improved, the applicability and generalization ability of the model in a small sample data scene are enhanced, and powerful support is provided for risk management and decision making of a credit institution.

Description

Technical Field

[0001] The present invention relates to the technical field of financial credit risk control, and in particular to a credit risk control modeling method and system. Background Art

[0002] In financial credit business, accurately assessing the credit risk of borrowers is the key to risk control. In order to effectively manage risks, credit institutions usually use a credit scoring model to score borrowers. The credit scoring model predicts the possibility of default by analyzing various characteristics of borrowers (such as income level, credit record, repayment history, etc.), thereby providing support for credit decisions.

[0003] Traditional scoring card model construction relies on a large amount of historical data to ensure that the model has good generalization ability and prediction accuracy. However, in some scenarios, such as new market expansion, new product launches, or analysis of specific customer groups, credit institutions may face the problem of insufficient historical data, resulting in restrictions on the construction and application of scoring card models. This small sample problem may cause the model to overfit during the training process and fail to effectively generalize to new data, which in turn affects the accuracy of risk assessment and the scientific nature of credit decisions.

[0004] At present, common methods to solve the small sample problem include data augmentation and transfer learning, but these methods have certain limitations. Although data augmentation methods can increase the sample size, they are prone to introduce noise and reduce the reliability of the model. Transfer learning requires a large amount of high-quality annotated data in similar fields, which is often difficult to meet in practical applications.

[0005] Therefore, the present invention provides a credit risk control modeling method and system. Summary of the invention

[0006] The present invention provides a credit risk control modeling method and system, which is used to generate multiple representative data sets through a preset method (i.e., Bootstrap sampling method), and train the risk control model on these data sets respectively, to ensure that the model can maintain stable performance under different data samples. This can not only reduce the random error caused by sample bias in a single model, but also improve the generalization ability of the model, so that it can better adapt to different credit business scenarios; in the iterative training and evaluation process, focus on the evaluation indicators of the model (i.e., key evaluation indicators such as AUC value, Divergence value and IV value) to ensure the model's ability to identify high-risk borrowers. Through multiple iterative calculations and evaluations, the performance of the model in distinguishing borrowers with different credit risk levels is optimized, the accuracy of identifying high-risk customers is significantly improved, and credit risk is reduced; by determining the initial model, the overfitting problem in small sample scenarios is effectively solved, and the target intermediate model obtained by each iterative training is trained according to a preset integrated learning algorithm to obtain the final credit risk control model, thereby effectively ensuring the comprehensive prediction ability and stability of the final credit risk control model. The present invention effectively improves the accuracy and stability of credit risk assessment, enhances the applicability and generalization ability of the model in small sample data scenarios, and provides strong support for risk management and decision-making of credit institutions.

[0007] A credit risk control modeling method, comprising:

[0008] Step 1: Obtain the business scenario requirements for credit risk control modeling, and determine the risk control modeling data set, initial model, evaluation indicators, and convergence logic based on the business scenario requirements;

[0009] Step 2: Train the initial model based on the risk control modeling data set, obtain the intermediate model, and quantify the evaluation indicators;

[0010] Step 3: When the quantitative calculation result of the evaluation index does not meet the convergence logic, the intermediate model is iteratively calculated according to the preset method, and when the iterative calculation reaches the convergence logic, the iterative calculation is stopped;

[0011] Step 4: Train the target intermediate model obtained from each iterative training according to the preset ensemble learning algorithm to obtain the final credit risk control model.

[0012] Preferably, a credit risk control modeling method, in step 1, obtains the business scenario requirements of credit risk control modeling, and determines the risk control modeling data set according to the business scenario requirements, including:

[0013] Reading the business scenario requirements of the credit risk control modeling, determining the scenario type of the business scenario requirements, and generating a first data extraction tag according to the scenario type;

[0014] Determine the data extraction scope according to the business scenario requirements, and generate a second data extraction tag according to the data extraction scope;

[0015] The first data extraction tag and the second data extraction tag are used to extract data in a preset database to obtain a risk control modeling data set.

[0016] Preferably, a credit risk control modeling method extracts data from a preset database using a first data extraction tag and a second data extraction tag to obtain a risk control modeling data set, including:

[0017] Extracting the label from the first data, performing type positioning in a preset database, and obtaining a first target data set consistent with the scene type according to the type positioning result;

[0018] Performing range positioning in the first target data set based on the second data extraction tag, and obtaining a second target data set consistent with the data extraction range according to the range positioning result;

[0019] The second target data set is extracted, wherein the second target data set is the risk control modeling data set.

[0020] Preferably, a credit risk control modeling method, in step 1, determines the initial model, evaluation indicators and convergence logic according to business scenario requirements, including:

[0021] Read the business scenario requirements, obtain the scenario characteristics of the business scenario requirements, and match the initial model in the model library according to the scenario characteristics;

[0022] Determine multiple evaluation indicators for evaluating the initial model according to business scenario requirements, select multiple evaluation indicators according to business scenario requirements, and set the convergence logic comprehensively based on the indicator characteristics of multiple evaluation indicators.

[0023] Preferably, a credit risk control modeling method, in step 3, when the quantitative calculation result of the evaluation index does not meet the convergence logic, the intermediate model is iteratively calculated according to a preset method, and when the iterative calculation reaches the convergence logic, the iterative calculation is stopped, including:

[0024] S301: Resample the risk control modeling data set based on a preset method to obtain target sample data;

[0025] S302: training the intermediate model according to the target sample data;

[0026] S303: selecting additional sample data other than the target sample data from the risk control modeling data set, and performing quantitative calculation on the evaluation index according to the additional sample data to obtain the evaluation index value;

[0027] S304 determines whether the evaluation index value reaches the convergence logic;

[0028] S305: If the evaluation index value does not reach the convergence logic, repeat steps S301-S304 to iteratively calculate the intermediate model until the evaluation index value reaches the convergence logic;

[0029] S306: If the evaluation index value reaches the convergence logic, the iterative calculation is stopped.

[0030] Preferably, in a credit risk control modeling method, in step 4, the target intermediate model obtained by each iterative training is trained according to a preset ensemble learning algorithm to obtain a final credit risk control model, including:

[0031] Reading a preset integrated learning algorithm and determining an integration rule of the preset integrated algorithm;

[0032] Combine the target intermediate models obtained from each iterative training according to the integration rule;

[0033] Train the combined target intermediate model according to the risk control modeling data set, and record the training results in real time;

[0034] Match training results with pre-set standards;

[0035] When the training result does not meet the preset standard, the combined target intermediate model is optimized until it meets the preset standard;

[0036] When the training results reach the preset standards, the final credit risk control model is obtained.

[0037] Preferably, a credit risk control modeling method, in step 4, after obtaining the final credit risk control model, includes:

[0038] Obtaining the model type of the final credit risk control model, and matching the first model packaging standard in the model packaging standard set according to the model type;

[0039] Obtaining a deployment platform for the final credit risk control model, and obtaining a standard state of the model in the deployment platform, and adjusting the first model packaging standard according to the standard state of the model in the deployment platform to obtain a second model packaging standard;

[0040] Encapsulating the final credit model according to the second model encapsulation standard to obtain a model package;

[0041] The model package is transferred to the deployment platform, and the model package is unpacked and deployed according to the deployment platform.

[0042] A credit risk control modeling system, comprising:

[0043] The risk control modeling initialization module is used to obtain the business scenario requirements of credit risk control modeling, and determine the risk control modeling data set, initial model, evaluation indicators and convergence logic according to the business scenario requirements;

[0044] The evaluation module is used to train the initial model based on the risk control modeling data set, obtain the intermediate model, and perform quantitative calculations on the evaluation indicators;

[0045] The risk control modeling iteration module is used to iteratively calculate the intermediate model according to the preset method when the quantitative calculation results of the evaluation indicators do not meet the convergence logic, and stop the iterative calculation when the iterative calculation reaches the convergence logic;

[0046] The risk control modeling integrated learning module is used to train the target intermediate model obtained from each iterative training according to the preset integrated learning algorithm to obtain the final credit risk control model.

[0047] Preferably, a credit risk control modeling system, a risk control modeling initialization module, comprises:

[0048] A label generation unit, for:

[0049] Reading the business scenario requirements of the credit risk control modeling, determining the scenario type of the business scenario requirements, and generating a first data extraction tag according to the scenario type;

[0050] Determine the data extraction scope according to the business scenario requirements, and generate a second data extraction tag according to the data extraction scope;

[0051] The data extraction unit is used to extract data from a preset database using the first data extraction tag and the second data extraction tag to obtain a risk control modeling data set.

[0052] Preferably, a credit risk control modeling system, a data extraction unit, comprises:

[0053] A first positioning subunit, configured to perform type positioning in a preset database according to the first data extraction tag, and obtain a first target data set consistent with the scene type according to the type positioning result;

[0054] A second positioning subunit is used to perform range positioning in the first target data set based on the second data extraction tag, and obtain a second target data set consistent with the data extraction range according to the range positioning result;

[0055] The extraction subunit is used to extract the second target data set, wherein the second target data set is the risk control modeling data set.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] A plurality of representative data sets are generated by a preset method (i.e., Bootstrap sampling method), and the risk control model is trained on these data sets respectively to ensure that the model can maintain stable performance under different data samples. In this way, not only can the random error caused by sample bias of a single model be reduced, but also the generalization ability of the model can be improved, so that it can better adapt to different credit business scenarios; in the process of iterative training and evaluation, the evaluation indicators of the model (i.e., key evaluation indicators such as AUC value, Divergence value and IV value) are focused on to ensure the model's ability to identify high-risk borrowers. Through multiple iterative calculations and evaluations, the performance of the model in distinguishing borrowers with different credit risk levels is optimized, the accuracy of identifying high-risk customers is significantly improved, and credit risk is reduced; by determining the initial model, the overfitting problem in the small sample scenario is effectively solved, and the target intermediate model obtained by each iterative training is trained according to the preset integrated learning algorithm to obtain the final credit risk control model, thereby effectively ensuring the comprehensive prediction ability and stability of the final credit risk control model obtained, the present invention effectively improves the accuracy and stability of credit risk assessment, enhances the applicability and generalization ability of the model in the small sample data scenario, and provides strong support for the risk management and decision-making of credit institutions.

[0058] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in this application document.

[0059] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0061] Figure 1 This is a flow chart of a credit risk control modeling method in an embodiment of the present invention;

[0062] Figure 2 This is a flowchart of step 1 in a credit risk control modeling method in an embodiment of the present invention;

[0063] Figure 3 The figure is a structural diagram of a credit risk control modeling system in an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0065] Embodiment 1:

[0066] This embodiment provides a credit risk control modeling method, such as Figure 1 As shown, including:

[0067] Step 1: Obtain the business scenario requirements for credit risk control modeling, and determine the risk control modeling data set, initial model, evaluation indicators, and convergence logic based on the business scenario requirements;

[0068] Step 2: Train the initial model based on the risk control modeling data set, obtain the intermediate model, and quantify the evaluation indicators;

[0069] Step 3: When the quantitative calculation result of the evaluation index does not meet the convergence logic, the intermediate model is iteratively calculated according to the preset method, and when the iterative calculation reaches the convergence logic, the iterative calculation is stopped;

[0070] Step 4: Train the target intermediate model obtained from each iterative training according to the preset ensemble learning algorithm to obtain the final credit risk control model.

[0071] In this embodiment, the business scenario requirements can be different links of the credit business, such as loan approval, quota setting, post-loan management and other scenarios, and what specific goals and expectations are there for accurately assessing the borrower's credit risk and predicting the possibility of default. For example, the business party may require the model to accurately identify high-risk customers to reduce the bad debt rate, or to reasonably determine the credit limit based on different customer characteristics. It may also include requirements for the accuracy, timeliness, and interpretability of the model, as well as special requirements arising from factors such as specific industries, market environments, and customer groups. Only by obtaining these business scenario requirements can we better build a credit risk control model that meets actual business needs and is effective.

[0072] In this embodiment, the selected data set should meet different sample size requirements according to the specific characteristics of the credit product. Generally, the data set sample size required for personal credit products should be larger than that for small and micro enterprise loan products. Generally, the data set should contain at least 50 good samples and 50 bad samples to ensure that the risk characteristics of the samples are statistically significant.

[0073] In this embodiment, the convergence logic may include convergence parameters pre-set according to business experience, such as indicator difference, maximum iteration rounds, etc.

[0074] In this embodiment, simple models are given priority in risk control model selection, such as Divergence scorecard, logistic regression and decision tree, to prevent overly complex models from overfitting in small sample scenarios.

[0075] In this embodiment, in the selection of evaluation indicators, the model loss function value (such as the Divergence value used in the scoring card model), AUC value and other indicators can be selected to comprehensively set the iteration stop logic. In the small sample data scenario, special attention is paid to the AUC-ROC indicator of the model to evaluate the model's ability to distinguish borrowers with different risk levels. At the same time, in the selection of evaluation indicators, the default probability predicted by the risk control model can be converted into a credit score, and the IV value can be calculated using this score and the sample label to evaluate the model's ability to distinguish different credit risk groups. Among them, the IV value is an indicator of the effectiveness of a variable in distinguishing different credit risk groups.

[0076] In this embodiment, the intermediate model can be an intermediate model obtained after training the initial model based on the risk control modeling data set.

[0077] In this embodiment, the preset method may be a Bootstrap method.

[0078] In this embodiment, the model obtained by each iterative training is trained using an ensemble learning algorithm to obtain a final model, including: model parameter average calculation, bagging ensemble learning and voting learning (ie, a preset ensemble learning algorithm).

[0079] In this embodiment, in step 2, an initial model is trained according to the risk control modeling data set to obtain an intermediate model, and the evaluation index is quantitatively calculated, including:

[0080] Obtain the obtained risk control modeling data set, and determine the outlier data in the risk control modeling data set based on the target value of the risk control modeling data set;

[0081] Clean the outlier data, and determine the mean of the risk control modeling data set based on the cleaning result and the target value of the risk control modeling data set, and fill the location of the cleaned outlier data based on the mean value;

[0082] The risk control modeling data set after data filling is standardized based on a preset value range, and the data features of the risk control modeling data set are extracted based on the standardized processing results;

[0083] Perform clustering processing on the risk control modeling data set based on data features to obtain the data category corresponding to the risk control modeling data set and the amount of risk control modeling data contained in each data category. When it is determined based on the amount of risk control modeling data that there is a sample deviation in the data category, sample the risk control modeling data in each data category to obtain a standard risk control modeling data set.

[0084] Perform structural analysis on the initial model to obtain the hierarchical structure in the initial model, and then perform structural decomposition on the initial model based on the hierarchical structure;

[0085] Extract the model parameters of each level based on the structural splitting results, and determine the model weights of different hierarchical structures based on the model parameters;

[0086] Based on the standard risk control modeling data set, each hierarchical structure in the initial model is iterated for the target number of times in descending order of model weight to obtain an intermediate model, and the frequency of regular traversal of the training results of each hierarchical structure is determined;

[0087] The training results are periodically traversed based on the regular traversal frequency, and the training results corresponding to each hierarchical structure are analyzed based on the periodic traversal results to obtain the quantitative value of the evaluation index of each hierarchical structure;

[0088] Based on the model weights of different hierarchical structures, the quantitative values ​​of the same category of evaluation indicators of different hierarchical structures are weighted averaged to obtain the quantitative value of the target evaluation indicator of the intermediate model.

[0089] The above-mentioned risk control modeling data set may be sample data obtained for model training.

[0090] The above target values ​​may be the specific values ​​corresponding to the data in the risk control modeling data set.

[0091] The above-mentioned outlier data may be data with abnormal values ​​in the risk control modeling data set, for example, data with values ​​exceeding the average value.

[0092] The above data filling can be to re-add data and assign values ​​to the locations where the outlier data are located through the mean value of the risk control modeling data set, in order to ensure the integrity of the risk control modeling data set.

[0093] The above-mentioned preset value range is set in advance, for example, it can be [0, 1], which is the basis for standardizing the risk control modeling data set. Standardization is to normalize the values ​​of different data in the risk control modeling data set to the preset value range, so as to facilitate the determination of the characteristics of the risk control modeling data set.

[0094] The above-mentioned data features may be the data value distribution and data structure of the risk control modeling data set.

[0095] The above-mentioned sample bias may occur when the amount of risk control modeling data contained in each data category is inconsistent, that is, sample bias exists.

[0096] The above-mentioned standard risk control modeling data set may be the result obtained by re-determining the risk control modeling data of each data category when there is a sample deviation, that is, the amount of data in each data category is the same.

[0097] The above-mentioned hierarchical structure may be the structural condition in the initial model, thereby facilitating effective training of the initial model according to the structural condition of the initial model.

[0098] The above model weights may be the importance of different hierarchical structures in the initial model in the entire model.

[0099] The above-mentioned periodic traversal frequency may be the number of times the training results of each hierarchical structure are checked within a certain period of time.

[0100] The working principle and beneficial effects of the above technical solution are: by analyzing and processing the obtained risk control modeling data set, the outlier data existing in the risk control modeling data set can be accurately and effectively proposed, and the risk control modeling data set can be effectively split into categories, thereby providing reliable data support for training the initial model; secondly, the initial model is trained by processing the obtained standard risk control modeling data set, and the results of the training process are cyclically traversed; finally, the cycle traversal results are processed and analyzed to achieve accurate and effective determination of the quantitative values ​​of the target evaluation indicators of the intermediate model, which provides strong support for the risk management and decision-making of credit institutions.

[0101] The working principle and beneficial effects of the above technical solution also include: generating multiple representative data sets through a preset method (i.e., Bootstrap sampling method), and training the risk control model on these data sets to ensure that the model can maintain stable performance under different data samples. This can not only reduce the random error caused by sample bias in a single model, but also improve the generalization ability of the model, making it better adapted to different credit business scenarios; in the iterative training and evaluation process, focus on the evaluation indicators of the model (i.e., key evaluation indicators such as AUC value, Divergence value, and IV value) to ensure the model's ability to identify high-risk borrowers. Through multiple iterative calculations and evaluations, the performance of the model in distinguishing borrowers with different credit risk levels is optimized, the accuracy of identifying high-risk customers is significantly improved, and credit risk is reduced; by determining the initial model, the overfitting problem in small sample scenarios is effectively solved, and the target intermediate model obtained by each iterative training is trained according to a preset integrated learning algorithm to obtain the final credit risk control model, thereby effectively ensuring the comprehensive prediction ability and stability of the final credit risk control model. The present invention effectively improves the accuracy and stability of credit risk assessment, enhances the applicability and generalization ability of the model in small sample data scenarios, and provides strong support for risk management and decision-making of credit institutions.

[0102] Embodiment 2:

[0103] Based on Example 1, this example provides a credit risk control modeling method, such as Figure 2As shown, in step 1, the business scenario requirements for credit risk control modeling are obtained, and the risk control modeling data set is determined according to the business scenario requirements, including:

[0104] S101: Reading business scenario requirements for credit risk control modeling, determining the scenario type of the business scenario requirements, and generating a first data extraction tag according to the scenario type;

[0105] S102: Determine a data extraction scope according to business scenario requirements, and generate a second data extraction tag according to the data extraction scope;

[0106] S103: Extract data using the first data extraction tag and the second data extraction tag in a preset database to obtain a risk control modeling data set.

[0107] In this embodiment, the first data extraction tag may be a data extraction identifier determined according to the scenario type after reading the business scenario requirements, and is used to limit the scenario type.

[0108] In this embodiment, the data extraction range may be a standard that is required to be followed when extracting training samples from a preset database.

[0109] In this embodiment, the second data extraction tag may be a data extraction identifier generated according to the data extraction range after being read according to business scenario requirements, and is used to limit the data range.

[0110] The working principle and beneficial effect of the above technical solution are: by obtaining the first data extraction label and the second data label, the risk control modeling data set can be effectively extracted, thereby improving the effectiveness and accuracy of the risk control modeling data set extraction.

[0111] Embodiment 3:

[0112] Based on Example 2, this example provides a credit risk control modeling method, which extracts data from a preset database using a first data extraction tag and a second data extraction tag to obtain a risk control modeling data set, including:

[0113] Extracting the label from the first data, performing type positioning in a preset database, and obtaining a first target data set consistent with the scene type according to the type positioning result;

[0114] Performing range positioning in the first target data set based on the second data extraction tag, and obtaining a second target data set consistent with the data extraction range according to the range positioning result;

[0115] The second target data set is extracted, wherein the second target data set is the risk control modeling data set.

[0116] In this embodiment, the type positioning may be determined according to the data type of the first data extraction tag in the preset database.

[0117] In this embodiment, the first target data set may be all data in a preset database having the same tag as the first data extraction tag.

[0118] In this embodiment, the preset database may be set in advance and include different types of risk control sample data.

[0119] In this embodiment, range positioning may be to limit the data value range in the first target data set according to the second data extraction tag, in order to lock in data that meets the value requirements.

[0120] The beneficial effect of the above technical solution is that by effectively locating the type and range of data in a preset database according to the first data extraction label and the second data extraction label, the effectiveness and accuracy of the risk control modeling data set extraction are effectively improved.

[0121] Embodiment 4:

[0122] Based on Example 1, this example provides a credit risk control modeling method. In step 1, an initial model, evaluation indicators, and convergence logic are determined according to business scenario requirements, including:

[0123] Read the business scenario requirements, obtain the scenario characteristics of the business scenario requirements, and match the initial model in the model library according to the scenario characteristics;

[0124] Determine multiple evaluation indicators for evaluating the initial model according to business scenario requirements, select multiple evaluation indicators according to business scenario requirements, and set convergence logic based on the indicator characteristics of multiple evaluation indicators.

[0125] In this embodiment, the scenario characteristics may be the scenario type and scenario application requirements corresponding to the business scenario requirements.

[0126] In this embodiment, the evaluation index may be applicable to rules or standards for performing performance testing on the initial model.

[0127] In this embodiment, the indicator feature may be the evaluation standard and the limited range defined for each evaluation indicator.

[0128] In this embodiment, the model library may be set in advance.

[0129] In this embodiment, simple models are preferentially matched in the initial model determination, such as Divergence scorecard, logistic regression and decision tree.

[0130] The working principle and beneficial effects of the above technical solution are: by reading the business scenario requirements, the scenario characteristics of the business scenario requirements can be accurately and effectively determined; secondly, the initial model is matched from the model library according to the scenario characteristics; at the same time, multiple evaluation indicators are determined according to business requirements; finally, the convergence logic is effectively determined according to the indicator characteristics of multiple evaluation indicators, which provides a reliable guarantee for credit risk control modeling.

[0131] Embodiment 5:

[0132] Based on Example 1, this example provides a credit risk control modeling method. In step 3, when the quantitative calculation result of the evaluation index does not meet the convergence logic, the intermediate model is iteratively calculated according to a preset method, and when the iterative calculation reaches the convergence logic, the iterative calculation is stopped, including:

[0133] S301: Resample the risk control modeling data set based on a preset method to obtain target sample data;

[0134] S302: training the intermediate model according to the target sample data;

[0135] S303: selecting additional sample data other than the target sample data from the risk control modeling data set, and performing quantitative calculation on the evaluation index according to the additional sample data to obtain the evaluation index value;

[0136] S304 determines whether the evaluation index value reaches the convergence logic;

[0137] S305: If the evaluation index value does not reach the convergence logic, repeat steps S301-S304 to iteratively calculate the intermediate model until the evaluation index value reaches the convergence logic;

[0138] S306: If the evaluation index value reaches the convergence logic, the iterative calculation is stopped.

[0139] In this embodiment, the target sample data may be the result obtained by resampling the risk control modeling data set using a preset method.

[0140] In this embodiment, the additional sample data may be data selected from the risk control modeling data set that is inconsistent with the target sample data.

[0141] In this embodiment, when resampling is performed, it is necessary to ensure that the probability of each sample being extracted is linearly related to the sample weight.

[0142] In this embodiment, let the selected evaluation index be X. After iterative calculation for n times, the calculation As the evaluation indicator of the current training round.

[0143] The working principle and beneficial effects of the above technical solution are: by using the Bootstrap method (i.e., the preset method) to resample from all the data to obtain a new data set, it is ensured that the probability of each sample being extracted is linearly related to the sample weight to maintain the representativeness and diversity of the data set, and the size of the extracted sample set is equal to the size of the original small sample data set to ensure that the sample credit risk characteristics reflected by each data set have the same statistical distribution, thereby ensuring that the evaluation index value meets the convergence logic and ensuring the accuracy of the construction of the credit risk control model.

[0144] Embodiment 6:

[0145] Based on Example 1, this example provides a credit risk control modeling method. In step 4, the target intermediate model obtained by each iterative training is trained according to a preset ensemble learning algorithm to obtain a final credit risk control model, including:

[0146] Reading a preset integrated learning algorithm and determining an integration rule of the preset integrated algorithm;

[0147] Combine the target intermediate models obtained from each iterative training according to the integration rule;

[0148] Train the combined target intermediate model according to the risk control modeling data set, and record the training results in real time;

[0149] Match training results with pre-set standards;

[0150] When the training result does not meet the preset standard, the combined target intermediate model is optimized until it meets the preset standard;

[0151] When the training results reach the preset standards, the final credit risk control model is obtained.

[0152] In this embodiment, the preset ensemble learning algorithm is set in advance.

[0153] In this embodiment, the preset standard is known in advance and is a standard used to measure whether the final result meets the requirements.

[0154] In this embodiment, the combined intermediate model may be optimized based on the training results and in combination with preset standards, and the parameters, weights, etc. of the combined intermediate model may be adjusted, and the adjustment values ​​are determined based on the actual training process.

[0155] In this embodiment, the target model obtained by each iterative training is trained using an ensemble learning algorithm to obtain the final credit risk control model. In the field of credit risk control modeling, the most frequently used and widely used models are scorecards and logistic regression based on WoE coding, both of which are linear models. Therefore, the ensemble learning step can be simplified, and the model parameters obtained by each iterative training are simply averaged to obtain the model parameters of the final model. The model parameters obtained in this way are more robust, so that the model scoring can more accurately reflect the actual credit status of the borrower and avoid misjudgment caused by overfitting.

[0156] The working principle and beneficial effect of the above technical solution are: the intermediate models obtained after each iterative training are combined and trained by a preset integrated learning algorithm, and the combined training results are verified to ensure the accuracy and reliability of the final credit risk control model.

[0157] Embodiment 7:

[0158] Based on Example 1, this example provides a credit risk control modeling method. In step 4, after the final credit risk control model is obtained, the method includes:

[0159] Obtaining the model type of the final credit risk control model, and matching the first model packaging standard in the model packaging standard set according to the model type;

[0160] Obtaining a deployment platform for the final credit risk control model, and obtaining a standard state of the model in the deployment platform, and adjusting the first model packaging standard according to the standard state of the model in the deployment platform to obtain a second model packaging standard;

[0161] Encapsulating the final credit model according to the second model encapsulation standard to obtain a model package;

[0162] The model package is transferred to the deployment platform, and the model package is unpacked and deployed according to the deployment platform.

[0163] In this embodiment, the model packaging standard set may be a set of standards for packaging different types of credit risk control models.

[0164] In this embodiment, the first model packaging standard may be a standard suitable for packaging the current credit risk control model.

[0165] In this embodiment, the standard state may be the operating status corresponding to the normal operation of the final credit risk control model in the deployment platform.

[0166] In this embodiment, the second model packaging standard may be a result obtained by adjusting the first model packaging standard according to the standard state.

[0167] The working principle and beneficial effects of the above technical solution are: according to the model type, the first model packaging standard can be effectively matched in the model packaging standard set, and by obtaining the standard state of the model in the deployment platform, the adjustment of the first model packaging standard can be effectively realized, which is beneficial to ensure the reliability of the packaging of the final credit model and the adaptability to the deployment platform, and then the packaging of the final credit model is effectively realized through the second model packaging standard. At the same time, according to the packaging and deployment of the model package on the deployment platform, the final credit model is effectively realized in a real business environment.

[0168] Embodiment 8:

[0169] This embodiment provides a credit risk control modeling system. Figure 3 As shown, including:

[0170] The risk control modeling initialization module is used to obtain the business scenario requirements of credit risk control modeling, and determine the risk control modeling data set, initial model, evaluation indicators and convergence logic according to the business scenario requirements;

[0171] The evaluation module is used to train the initial model based on the risk control modeling data set, obtain the intermediate model, and perform quantitative calculations on the evaluation indicators;

[0172] The risk control modeling iteration module is used to iteratively calculate the intermediate model according to the preset method when the quantitative calculation results of the evaluation indicators do not meet the convergence logic, and stop the iterative calculation when the iterative calculation reaches the convergence logic;

[0173] The risk control modeling integrated learning module is used to train the target intermediate model obtained from each iterative training according to the preset integrated learning algorithm to obtain the final credit risk control model.

[0174] The working principle and beneficial effects of the above technical solution are: generate multiple representative data sets through a preset method (i.e., Bootstrap sampling method), and train the risk control model on these data sets to ensure that the model can maintain stable performance under different data samples. This can not only reduce the random error caused by sample bias in a single model, but also improve the generalization ability of the model, so that it can better adapt to different credit business scenarios; in the iterative training and evaluation process, focus on the evaluation indicators of the model (i.e., key evaluation indicators such as AUC value, Divergence value and IV value) to ensure the model's ability to identify high-risk borrowers. Through multiple iterative calculations and evaluations, the performance of the model in distinguishing borrowers with different credit risk levels is optimized, the accuracy of identifying high-risk customers is significantly improved, and credit risk is reduced; by determining the initial model, the overfitting problem in small sample scenarios is effectively solved, and the target intermediate model obtained by each iterative training is trained according to a preset integrated learning algorithm to obtain the final credit risk control model, thereby effectively ensuring the comprehensive prediction ability and stability of the final credit risk control model. The present invention effectively improves the accuracy and stability of credit risk assessment, enhances the applicability and generalization ability of the model in small sample data scenarios, and provides strong support for risk management and decision-making of credit institutions.

[0175] Embodiment 9:

[0176] Based on Example 8, this embodiment provides a credit risk control modeling system, a risk control modeling initialization module, including:

[0177] A label generation unit, for:

[0178] Reading the business scenario requirements of the credit risk control modeling, determining the scenario type of the business scenario requirements, and generating a first data extraction tag according to the scenario type;

[0179] Determine the data extraction scope according to the business scenario requirements, and generate a second data extraction tag according to the data extraction scope;

[0180] The data extraction unit is used to extract data from a preset database using the first data extraction tag and the second data extraction tag to obtain a risk control modeling data set.

[0181] The working principle and beneficial effect of the above technical solution are: by obtaining the first data extraction label and the second data label, the risk control modeling data set can be effectively extracted, thereby improving the effectiveness and accuracy of the risk control modeling data set extraction.

[0182] Embodiment 10:

[0183] Based on Example 9, this embodiment provides a credit risk control modeling system, a data extraction unit, including:

[0184] A first positioning subunit, configured to perform type positioning in a preset database according to the first data extraction tag, and obtain a first target data set consistent with the scene type according to the type positioning result;

[0185] A second positioning subunit is used to perform range positioning in the first target data set based on the second data extraction tag, and obtain a second target data set consistent with the data extraction range according to the range positioning result;

[0186] The extraction subunit is used to extract the second target data set, wherein the second target data set is the risk control modeling data set.

[0187] In the above technical solution, the first positioning subunit is used to find data matching a specific scene type in a preset database using the first data extraction tag. "Type positioning" means that it filters data according to the business type or feature type of the data. In operation, the data tags are first identified, and then the data matching these tags is searched in the database, and finally a data set consistent with the scene type is obtained.

[0188] Second positioning subunit: Based on the data set extracted by the first positioning subunit, the second positioning subunit further narrows the data scope using the second data extraction tag. Here, "scope positioning" refers to filtering data based on time range, geographic range, or other business-related ranges. Limit the scope of the first target data set to ensure that the extracted data meets specific business needs or analysis goals.

[0189] The function of the extraction subunit is to extract the data set after two positioning and screening. This data set is the data set used for risk control modeling.

[0190] In actual implementation, the first positioning subunit is pre-configured with a Label-Studio annotation component for automatic annotation of scene types. The Label-Studio annotation component is a annotation platform. In actual implementation, the Label-Studio annotation component sets the annotation identifier through a pre-set specific scene type, and automatically performs type positioning annotation when any annotation identifier that meets the specific production type is identified; then the second positioning subunit performs association verification between the annotation identifier and the business requirement, and when the association verification is met, range positioning is achieved, and the dog issues an alarm to improve safety.

[0191] The beneficial effect of the above technical solution is: effectively improving the effectiveness and accuracy of risk control modeling data set extraction.

[0192] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A credit risk control modeling method, characterized in that: include: Step 1: Obtain the business scenario requirements for credit risk control modeling, and determine the risk control modeling data set, initial model, evaluation indicators, and convergence logic based on the business scenario requirements; Step 2: Train the initial model based on the risk control modeling data set, obtain the intermediate model, and quantify the evaluation indicators; Step 3: When the quantitative calculation result of the evaluation index does not meet the convergence logic, the intermediate model is iteratively calculated according to the preset method, and when the iterative calculation reaches the convergence logic, the iterative calculation is stopped; Step 4: Train the target intermediate model obtained from each iterative training according to the preset ensemble learning algorithm to obtain the final credit risk control model.

2. A credit risk control modeling method according to claim 1, characterized in that: In step 1, obtain the business scenario requirements for credit risk control modeling, and determine the risk control modeling data set based on the business scenario requirements, including: Reading the business scenario requirements of the credit risk control modeling, determining the scenario type of the business scenario requirements, and generating a first data extraction tag according to the scenario type; Determine the data extraction scope according to the business scenario requirements, and generate a second data extraction tag according to the data extraction scope; The first data extraction tag and the second data extraction tag are used to extract data in a preset database to obtain a risk control modeling data set.

3. A credit risk control modeling method according to claim 2, characterized in that: The first data extraction tag and the second data extraction tag are used to extract data in a preset database to obtain a risk control modeling data set, including: Extracting the label from the first data, performing type positioning in a preset database, and obtaining a first target data set consistent with the scene type according to the type positioning result; Performing range positioning in the first target data set based on the second data extraction tag, and obtaining a second target data set consistent with the data extraction range according to the range positioning result; The second target data set is extracted, wherein the second target data set is the risk control modeling data set.

4. A credit risk control modeling method according to claim 1, characterized in that: In step 1, the initial model, evaluation indicators, and convergence logic are determined according to the business scenario requirements, including: Read the business scenario requirements, obtain the scenario characteristics of the business scenario requirements, and match the initial model in the model library according to the scenario characteristics; Determine multiple evaluation indicators for evaluating the initial model according to business scenario requirements, select multiple evaluation indicators according to business scenario requirements, and set convergence logic based on the indicator characteristics of multiple evaluation indicators.

5. A credit risk control modeling method according to claim 1, characterized in that: In step 3, when the quantitative calculation result of the evaluation index does not meet the convergence logic, the intermediate model is iteratively calculated according to the preset method, and when the iterative calculation reaches the convergence logic, the iterative calculation is stopped, including: S301: Resample the risk control modeling data set based on a preset method to obtain target sample data; S302: training the intermediate model according to the target sample data; S303: selecting additional sample data other than the target sample data from the risk control modeling data set, and performing quantitative calculation on the evaluation index according to the additional sample data to obtain the evaluation index value; S304 determines whether the evaluation index value reaches the convergence logic; S305: If the evaluation index value does not reach the convergence logic, repeat steps S301-S304 to iteratively calculate the intermediate model until the evaluation index value reaches the convergence logic; S306: If the evaluation index value reaches the convergence logic, the iterative calculation is stopped.

6. A credit risk control modeling method according to claim 1, characterized in that: In step 4, the target intermediate model obtained from each iterative training is trained according to the preset ensemble learning algorithm to obtain the final credit risk control model, including: Reading a preset integrated learning algorithm and determining an integration rule of the preset integrated algorithm; Combine the target intermediate models obtained from each iterative training according to the integration rule; Train the combined target intermediate model according to the risk control modeling data set, and record the training results in real time; Match training results with preset standards; When the training result does not meet the preset standard, the combined target intermediate model is optimized until it meets the preset standard; When the training results reach the preset standards, the final credit risk control model is obtained.

7. A credit risk control modeling method according to claim 1, characterized in that: In step 4, after obtaining the final credit risk control model, it includes: Obtaining the model type of the final credit risk control model, and matching the first model packaging standard in the model packaging standard set according to the model type; Obtaining a deployment platform for the final credit risk control model, and obtaining a standard state of the model in the deployment platform, and adjusting the first model packaging standard according to the standard state of the model in the deployment platform to obtain a second model packaging standard; Encapsulating the final credit model according to the second model encapsulation standard to obtain a model package; The model package is transferred to the deployment platform, and the model package is unpacked and deployed according to the deployment platform.

8. A credit risk control modeling system, characterized in that: include: The risk control modeling initialization module is used to obtain the business scenario requirements of credit risk control modeling, and determine the risk control modeling data set, initial model, evaluation indicators and convergence logic according to the business scenario requirements; The evaluation module is used to train the initial model based on the risk control modeling data set, obtain the intermediate model, and perform quantitative calculations on the evaluation indicators; The risk control modeling iteration module is used to iteratively calculate the intermediate model according to the preset method when the quantitative calculation results of the evaluation indicators do not meet the convergence logic, and stop the iterative calculation when the iterative calculation reaches the convergence logic; The risk control modeling integrated learning module is used to train the target intermediate model obtained from each iterative training according to the preset integrated learning algorithm to obtain the final credit risk control model.

9. A credit risk control modeling system according to claim 8, characterized in that: The risk control modeling initialization module includes: A label generation unit, for: Reading the business scenario requirements of the credit risk control modeling, determining the scenario type of the business scenario requirements, and generating a first data extraction tag according to the scenario type; Determine the data extraction scope according to the business scenario requirements, and generate a second data extraction tag according to the data extraction scope; The data extraction unit is used to extract data from a preset database using the first data extraction tag and the second data extraction tag to obtain a risk control modeling data set.

10. A credit risk control modeling system according to claim 9, characterized in that: Data extraction unit, comprising: A first positioning subunit, configured to perform type positioning in a preset database according to the first data extraction tag, and obtain a first target data set consistent with the scene type according to the type positioning result; A second positioning subunit is used to perform range positioning in the first target data set based on the second data extraction tag, and obtain a second target data set consistent with the data extraction range according to the range positioning result; The extraction subunit is used to extract the second target data set, wherein the second target data set is the risk control modeling data set.

Images