Predicting Customer Churn in Digital Banking Services Using Machine Learning
No Thumbnail Available
Date
2025-06
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
an increasingly competitive digital banking environment, retaining customers is a key consideration in emerging markets like Ethiopia, where the adoption of digital offerings today is increasing rapidly. This study solves the gap of the lack of Ethiopian digital-specific prediction models, and the limited incorporation of sentiment analysis and probabilistic approaches. It proposes a suite of machine learning–based predictions of the risk of customer churn in the Ethiopia digital banking environment utilizing a comprehensive (one year) dataset of behavioral transactions, demographic details, and customer sentiment measures based on surveys. The researchers trained, tested, and compared five supervised learning models: Logistic Regression, Decision Tree, Gradient Boosting, Random Forest, and. Neural Network.
The Random Forest approach performed best overall scoring 95% accuracy based on 0.204 score of log loss and ROC-AUC score of 0.984. The addition of sentiment features significantly improves the model's performance and highlights the potential value of obtaining customer sentiment responses on their likelihood of churn. Feature importance analysis using SHAP revealed that the most influential predictors of churn were EASY_SCORE, REGION, LAST_LOGIN, CONTINUE_SCORE, SECURE_SCORE, BALANCE, and NO_DB_TRN. The study also implemented individualized churn probability predictions. These results affirm the need for customer-centric churn models that account for both behavior and perception.
Overall, this study underscores the importance of blending behavioral analytics with customer feedback to develop proactive, personalized retention strategies in Ethiopia’s growing digital banking ecosystem.
Description
Keywords
Predicting Customer Churn, Digital Banking, Machine Learning