Application of machine learning in predicting credit card customer churn

Abstract

This paper aims to forecast the likelihood of customers leaving bank credit card services using machine learning methods. The methods used include Random Forest, SVM, Naïve Bayes, Logistic regression, and a combination of all four methods. The results show that those methods have good predictive quality with high accuracy. In particular, the prediction results by Random Forest are the best on all criteria from accuracy, sensitivity, specificity to F Score. In addition, the most important factors affecting the customer churn probability are indicators related to transaction history, products, and the relationship between the bank and the customer. This result can provide recommendations for bank managers in retaining customers who are using credit card services.