A Novel SHAP-Guided Bayesian Optimization Method for Telecom Customer Churn Prediction and Retention

Due to the increasing competition in the telecommunications industry and the continuous changes in the market environment, customer churn has become an increasingly serious problem. To help telecom companies accurately predict customer churn and develop effective retention strategies, A Bayesian Optimization-Infused Learning Model (XGBayes) is proposed. By optimizing hyperparameters and feature engineering, the goal is to improve prediction accuracy while enhancing the model's interpretability through SHAP analysis. Experiments conducted on customer data from a telecom company demonstrate that the proposed model outperforms existing models such as logistic regression (LG) random forest (RF), decision trees (DT), K-nearest neighbors (KNN) and traditional XGBoost in terms of prediction performance. SHAP analysis is further used to reveal the key drivers of customer churn and quantify the contribution of each feature to the prediction results. Through these analyses, more targeted customer retention strategies are provided to telecom operators, and effective decision support is offered for optimizing customer retention and reducing churn.