Hybrid Method for Churn Prediction Model in The Case of Telecommunication Companies

Internet Service Provider Companies are competing in offering attractive products to customers. The companies even mutually do acquisition the customers from other telecommunication companies. Subscribers who decided to unsubscribe the service from a service company is known as Churn. Churn rate continues to increase every year and the competition for internet service providers is getting tougher. The reason for the high Churn rate is that the retention incentive offer was rejected by Churn's potential customers. This is because the company was late in offering Churn retention incentives and customers had switched to competitors. One way to deal with this problem is to predict the possibility of customer's Churn. The purpose of this study is to create a Churn prediction model based on customer datasets and to develop a classification system that indicates customer will Churn or not. This study aims to develop and examined several hybrid models which are logistic regression with LASSO penalty, logistic regression with elastic net penalty, logistic regression with adaboost, and XGBoost with tuning parameters. The default methods for the classification process are logistic regression, Support Vector Machine (SVM), random forest and XGBoost. The results obtained from the 8 model prediction tests, where Churn predictions model using the hybrid method XGBoost with parameter tuning showed the best performance with Area Under Curve (AUC) 0.968, accuracy 94.3%, recall 98% and precision 96%.