Abstract: In order to improve the profits of enterprises and reduce operating costs, it is necessary to predict the loss of users, and carry out precise marketing in advance to retain users. The XGBoost model is established to train the user churn data, and the importance ranking of input features is obtained. The Top-K feature is selected to obtain a new training set. On the one hand, the XGBoost model of Bayesian optimization is established based on the training set, and the optimal parameters are found by Bayesian optimization; On the other hand, 8 models are selected to construct and verify the model, and the model is evaluated in precision, accuracy, recall and F₁ value respectively. Experimental results show that XGBoost model based on Bayesian optimization has better prediction results and higher efficiency than other models in telecom user churn prediction.

Key words: Bayesian optimization, XGBoost, user churn