Credit cards are a bank business in which high income and heavy risk coexist. Along with the de-velopment of the credit card business, banks are using the Internet and mobile data to establish customer credit rating system. How to evaluate customer credit from the information that cus-tomers fill in, and how to identify the information true or false, and what type of information that customers are asked to fill are crucial for banks. Based on the credit card customer data of 2005 in Taiwan, this article established Lasso-Logistic model and random forest model to explore the key factors which effect customer credit, including individual characteristics and some objective cha-racteristics. Through comparing the prediction accuracy of the model and F score index, we selected the model of better prediction effect to forecast the bank credit card defaults. The establishment of the credit card default prediction model and the exploration of the key factors influencing the customer credit not only have a important guidance value for banks to choose customers and design data, but also can provide certain theoretical support for the credit decisions. In addition, it has a strong theoretical and practical significance.