Bank customer churn prediction using a reproducible and explainable machine learning framework

Abstract

Predicting customer churn is crucial for banks because retaining existing customers is more cost-effective than acquiring new ones. This study presents a transparent and reproducible machine learning framework for predicting bank customer churn using a public dataset of 10,000 customers. Unlike previous work, the study resolves inconsistencies in the target variable and properly encodes categorical features such as Geography and Gender. Six machine learning models, including Logistic Regression, Decision Tree, Random Forest, KNearest Neighbor, Naive Bayes, and ensemble voting methods are evaluated using accuracy, precision, recall, and F1-score. The Bayesian Network model is applied with proper discretization of continuous features to ensure reproducibility. Explainable AI tools such as LIME and SHAP are used to interpret model predictions, providing insight into which features influence churn. The final model is integrated into a Flask-based interface for real-time prediction, offering banks a practical and reliable tool to reduce churn and improve customer retention. This framework emphasizes methodological rigor, reproducibility, and explainability, making it a novel and actionable approach to prediction churn in banking.