Predictive Banking: A Deep Ensemble Customer Churn Prediction Model For Enhanced Customer Retention

The issue of customer churn is increasingly critical for numerous or-ganizations, particularly within the banking sector, due to its negative impact on revenue. Factors such as market saturation, changing customer preferences, and rising competition contribute to this challenge. Identifying potential churners early enables banks to proactively engage with these customers, thereby improv-ing their CRM strategies to be more assertive and customer-centric. This paper introduces an innovative deep-learning ensemble model designed to predict cus-tomer churn in the banking industry. The proposed model operates on four levels. The initial two levels employ three essential data preprocessing techniques along-side a comprehensive feature reduction system utilizing XGBoost. In the third level, four distinct classifiers—CNN, SVM, LR, and DT—are trained as base classifiers. Finally, the predictions from these base classifiers are combined using a stacking ensemble method, which capitalizes on the strengths of each classifier by incorporating CatBoost as a meta-learner. Experimental results indicate that the proposed model surpasses other machine learning, deep learning, and ensem-ble algorithms, such as RF, SVM, CNN, and AdaBoost, achieving an impressive accuracy of 87.95%.