Client Churn Prediction in Retail Finance Institutions using Optimized XGBoost and Explainable AI Methods

Customer churn prediction has developed into a field of study. Experimental literature in banking industry has shown that the costs of retaining customers are significantly lower than customer acquisition costs are. As a result, the paper provides the proposal of a machine-learning model that would predict a possible churner at the initial stage based on the demographic, behavioral, and transactional data. The research includes the processing of data, features selection, and the implementation of the different predictive models, such as “Logistic Regression (LR)”, “random forest (RF)”, and “gradient-boosting methods (GB)”. In an effort to increase the level of interpretability, the framework adopts “SHAP” analysis to explain the most pronounced features that can be attributed to customer churn. The suggested system has the potential to be a utility to the banks and non-banking financial corporations (NBFCs), since it will enable the identification of high-risk customer groups and allow understanding which factors are major contributors to churn. This will allow the financial institutions to curb the churn levels within the high-value and target customers.