After more than ten years of rapid development, the bank credit card market is now very competitive. Many banks blindly pursue the increase in credit card issuance, ignoring the maintenance of existing customers, resulting in an increasing number of "dormant cards" and frequent customer loss, which is likely to become a hidden danger for the stable development of the banking industry. It is particularly important to Change the business philosophy of the banking industry and improve the loyalty of credit card customers for the development of the banking industry. Therefore, it is very important to establish a model based on LightGBM to predict the loss of credit card customers, and identify customers with churn tendencies in time, so as to better maintain customer resources. Through experimental analysis, it is found that the LightGBM model has the highest accuracy, reaching 95%. This also provides help for bank managers in customer churn prediction and customer retention, which has certain research significance.