A New Hybrid Churn Prediction Model Based on Logistic Regression and Extreme Gradient Boosting

Customer churn is regarded as one of the most important issues in customer relationship management (CRM).Although many traditional machine learning algorithms and statistical methods have been applied to churn prediction, it remains a challenge to build churn models on imbalance data sets.In this paper, we present a two-phase hybrid model based on logistic regression and extreme gradient boosting (XGBoost) technique to address this issue.The new model attempts to solve class imbalance issue by paying more attention to accurately classify the borderline instances.Comprehensive comparison experiments were carried out on four realworld data sets from the telecommunication industry.These results show that the hybrid model improves the predictive performance than logistic regression XGBoost and other benchmark models, especially in term of profit-best measure.The new model provides a useful guideline for churn prediction with class imbalance.