Customer Relationship Management: Bayesian-Based Xgboost Customer Churn Prediction and Model Interpretability

In today's dynamic economies, businesses and companies are working hard to attract customers. According to studies, the retention of existing consumers is 6 to 10 times cheaper than the addition of new customers. Therefore, customer churn has become a significant challenge for the industry,especially in saturated markets. Moreover, prediction methods face many challenges, such as imbalanced datasets, dirty data, low accuracy, etc. However, Ensemble methods show good performance in distinguishing churners and non-churners. In this paper, prevalent ensemble methodshave been implemented and compared. Also, Bayesian-based Xgboost is used for better performance. The results show that Bayesian-based Xgboost makes more accurate and better results in imbalanced evaluation metrics such as AUC and F-score. Moreover, in this paper, the interpretability of the machinelearning model and customer lifetime value of churners are discussed. So, Managers and decision makers could understand model outputs and how features impact the results and also make decent marketing decisions

Keywords:

Customer Churn Prediction , Ensemble Learning , Hyperparameter tuning , Xgboost , Interpretability , CLV