Application Research Of Credit Card User Credit Risk Prediction Model Based On CatBoost Algorithm

Internet technology has promoted the rapid development of credit card business.The credit card not only brought convenient consumer experience to users,but also brought tremendous benefits to banks of the issuance of credit card,but the credit card business also brought the corresponding risks.Credit risk in credit card business may bring greater losses to the card issuer.Therefore,how to identify customers with default risk and avoid credit risk has become an important part of credit card risk management.With the advent of the big data era,machine learning has begun to be applied in various fields,and it has become possible to apply machine learning algorithms to credit risk research.CatBoost algorithm has been applied to e-commerce,disease prediction and other fields to achieve better results,so this article will use CatBoost algorithm to study credit card credit risk.This article first preprocessed the data of 24 variables,including credit limit,gender,age,education,marital status,repayment amount,repayment status,and bills payable,and selected 19 of them as input to the model.Variables,establish a credit risk prediction model for credit card users based on the CatBoost algorithm.In order to verify the superiority of the CatBoost algorithm in credit card credit risk prediction,this paper compares the model established with the model based on Logistic regression,random forest,GBDT,and XGBoost classification algorithms,and analyzes and compares the accuracy and AUC value of each model.And ROC curve.The comparison results show that the accuracy of CatBoost among the five classification prediction models is 91.72%,the highest among the five models,and the AUC value is0.86.Among the remaining four models,the accuracy of XGBoost is the highest86.27%,and the AUC value is 0.78.The prediction accuracy of CatBoost is 5.54%higher than that of XGBoost,which verifies the superiority of the CatBoost algorithm in credit card credit risk prediction.Compared with the other four algorithms,the model based on the CatBoost algorithm has higher classification accuracy and can provide a reference for commercial banks' credit card risk prediction.Through the analysis of the influencing factors of credit card default,it is found that the default behavior of credit card users in the short term is mainly related to the recent bill amount.When the card issuer needs to predict the possibility of credit card user default in the short term,it is recommended to take the recent user's bill and repayment status as The main factors are analyzed and judged.By predicting the credit risk of credit card users,it is possible to identify customers with default risk and take relevant risk management measures to them,minimize the loss of credit risk to the credit card business,protect the legitimate interests of card issuers,and maintain the health of the credit card business development.