Predictive Modeling and Risk Scoring for Bank Customer Churn

Abstract

Customer churn is a significant challenge for retail banking institutions, as customer loss can adversely affect revenue stability, customer lifetime value, and opportunities for cross-selling and upselling. This study presents a machine-learning-based customer churn prediction and risk-scoring system using a European bank customer dataset comprising 10,000 records. The proposed workflow includes data preprocessing, exploratory data analysis, feature engineering, categorical encoding, model development, evaluation, and deployment through an interactive Streamlit application. Logistic Regression, Decision Tree, Random Forest, and Gradient Boosting were evaluated using Accuracy, Precision, Recall, F1-score, and ROC-AUC. Gradient Boosting achieved the best performance, with 87.10% accuracy and 86.69% ROC-AUC. Feature-importance analysis identified Age, Num Of Products, Is Active Member, Balance, and Geography_Germany as influential features. The deployed application generates customer-level churn probabilities and risk categories to support proactive, data-driven customer retention strategies.

KEYWORDS: Customer churn, Machine learning, Predictive analytics, Gradient boosting, Banking analytics.