Different ML-based strategies for customer churn prediction in banking sector

Churn prediction models are used to determine why and when customers are likely to discontinue their service in a variety of commercial fields, such as the banking and telecom industries. Machine learning methods are mostly utilized to construct churn models. This study proposes three models based on various feature set combinations and machine learning algorithms for predicting credit card customer churn. Model 1 takes into account every variable, Model 2 separates categorical and continuous features, and Model 3 uses feature selection techniques to focus on important features only. Different machine learning models including decision trees (C5.0 and CR Tree), Random Forest, Support Vector Machine (SVM), K Nearest Neighbour (KNN), XGBoost and Logistic Regression are applied on the models developed. The results assist us in determining the best machine learning models for churn prediction as well as the key features associated with churners. At the end, we present how well our models perform in comparison to a few other models.