Research On Loan Risk Control Model Based On Machine Learning

The traditional loan risk control technology has been unable to meet the current complex loan business needs.Financial institutions also need to improve the accuracy of loan prediction to reduce the bad debt rate.At the same time,the loan prediction model is required to have better interpretability,so that financial institutions can Strengthen the protection of their own rights and interests and provide lending reference for loan practitioners.Therefore,the main research work of this thesis is as follows:(1)Four common scenarios in financial risk control are analyzed to extract attributes with common characteristics as data input.The input data attributes are desensitized,missing values and outliers are processed,and data desensitization processing and data screening methods based on group stability indicators are proposed according to the needs of financial scenarios to improve model performance from the data source.(2)According to the needs of the accuracy and interpretability of the model in the financial risk control business,this thesis proposes that the CatBoost-LR fusion model based on Bagging improves the prediction accuracy by 5.31%compared to the single structure CatBoost.At the same time,attention and mask processing mechanisms are added to the rule-based representation learner method,which improves the robustness and prediction performance of the model.(3)According to the actual needs of loan practitioners,this thesis designs and implements a loan default prediction system.The system is implemented under the QT development framework,and the black-box testing method is used to test its functionality.The test proves that the system function meets the expected and actual needs.