Abstract: In recent years, banks have faced a loss of customers, called churn. This degrades the reputation of banks; hence, it is important to determine the difficulties that customers face. In business, churn prediction is helpful in determining metrics such as customer retention and revenue generation for various forms of Customer Relationship Management (CRM) techniques to forecast whether a customer will exit the bank. This study aims to develop a Profit-driven Churn Prediction (PCP) model using three Machine Learning (ML) techniques: an Artificial Neural Network (ANN), a Support Vector Machine (SVM), and a Random Forest Algorithm (RFA). The PCP model is developed using a correlation-based feature selection and a highly accurate ML classifier. This prediction model is trained using previous data to categorize consumers as non-churners or future churners. Customers' behavioral and demographic features are considered reliable indicators of churn prediction. The developed PCP model is tested using a bank customer churn prediction dataset obtained from the Kaggle repository. The RFA, SVM, and ANN algorithms achieved overall accuracies of 86%, 82.3%, and 97%, respectively, for the churn dataset. The classification accuracy serves as the basis for the performance of the ML classifier. Hence, the ANN classifier is more accurate than the other classifiers in this study; they have been employed in PCP for churn prediction.

Key words: customer churn, machine learning, support vector machine, artificial neural network, random forest algorithm