The development of communication technology and the rapid growth of the number of mobile network service users have made the competitive situation in the market of communication service increasingly fierce, and maintaining the stock of users is of great significance to the sustainable development of telecommunication enterprises. In this paper, we collect relevant data features of telecommunication users, and after pre-processing the features with RFM model, we use XGBoost model to analyze the importance of each user’s feature value. Then we use the secondary classification Stacking integration model that combines the base learner and the meta-learner to predict the telecom subscriber churn. Comparative validation reveals that the prediction model in this paper shows excellent prediction performance in all four datasets. Practical application results show that the effectiveness of churn maintenance efforts by telecom companies is improved after applying the model, and the average maintenance response rate reaches 50.63% in the first quarter of 2024. The prediction model proposed in this paper based on the binary classification method can assist telecommunication companies to manage the stock of subscribers, optimize the maintenance work plan, and reduce the subscriber churn rate in the telecommunication work period.