Predicting Risk Factors for Customer Churn with Explainable Artificial Intelligence
No Thumbnail Available
Date
2025-09-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Customer retention remains a critical challenge in the banking sector, particularly in emerging economies where customer behavior is rapidly evolving. Despite access to extensive data, Bunna Bank still relies on conventional customer relationship man agement practices that are often manual, reactive, and lacking analytical depth. This study focuses on predicting the risk factors for customer churn at Bunna Bank using machine learning and explainable artificial intelligence methods, using a real-world dataset of 308,293 customer records with 19 attributes having transactional behavior, account activity, and demographic information. Six machine learning classifiers were evaluated: Random Forest, Gradient Boosting, XGBoost, LightGBM, Extra
Trees, and Multi-Layer Perceptron. Before modeling, the dataset was balanced using SMOTE to address class imbalance, and a hybrid feature selection strategy was applied to identify the most relevant attributes. XGBoost achieved the highest accuracy (92.23%), along with strong precision (0.96), recall (0.94), F1-score (0.95), and AUC (0.97), making it the most effective model for identifying high-risk customers. To enhance interpretability, SHAP and LIME were applied to the XGBoost model, providing both global and local explanations. The analysis revealed that features such as Has Mobile Banking, Is USSD Active, Transaction Count, Transaction Recency, Age, and ATM Activity were the most influential drivers of churn risk. Active digital banking usage and recent transactions were protective factors, while younger customers with low engagement showed higher churn risk. A prototype was de
veloped using Streamlit to risk factor prediction that integrate predictive modeling with XAI enables accurate and transparent into customer churn
Description
Keywords
Customer Churn, Risk Factor Prediction, Machine Learning, Model Explainability