With the increasing competition pressure among tourism e-commerce platforms, how to reduce customer churn to the greatest extent is of great significance to tourism e-commerce platforms. Based on this, this paper takes Ctrip's hotel customer-related data as an example, and first uses a supervised feature selection method based on mutual information to select features that have an important impact on customer churn. Then, by using the cross-validation method, combined with evaluation indicators such as accuracy rate, F1-sorce, AUC, etc., select the best model set from Logistic regression, Support Vector Machine, Decision Tree, Random Forest, GBDT, XGBoost, and LightGBM. A subset of the optimal models, and then the optimal model fusion is performed. The empirical results show that the multi-model fusion has higher accuracy and stability. In addition, based on the model fusion results, this paper obtains the importance ranking of customer personal characteristics. Finally, this paper puts forward relevant suggestions on how to accurately manage Ctrip and reduce the customer churn rate.