In order to identify customers with default risk and avoid credit risk, the application of real-time data analysis and predictive maintenance in credit risk management of commercial banks has been proposed. This paper will use CatBoost algorithm to study credit risk of credit card. This paper first preprocesses the data of 24 real-time variables, such as credit line, gender, age, education, marital status, repayment amount, repayment status, and bills payable, and selects 19 of them as the input variables of the model to establish a credit card user credit risk prediction model based on CatBoost algorithm. The results show that the accuracy of CatBoost is 91.73%, which is the highest among the five models, and the accuracy of Logistic is 74.39%, which is the lowest among the five models. Compared with other algorithms, CatBoost algorithm has higher classification accuracy for credit default prediction of credit card users.