An Efficient Customer Churn Prediction Technique Using Combined Machine Learning in Commercial Banks

In this study, addressing the critical issue of early customer churn detection in the banking industry is acknowledged as pivotal for augmenting customer trust and retention. A stacked model, designed and structured across two levels, aims to enhance prediction accuracy. At Level 0, a combination of four distinct models, K-nearest neighbor, XGBoost, random forest, and support vector machine, is utilized, with each contributing unique analytical strengths. Level 1 is designed by the aggregation of through regression modeling, according to the way logistic regression, recurrent neural networks, and deep learning neural networks, thereby refining predictions. The stacked model’s efficacy is substantiated by superior performance metrics. Notably, the highest achievements are observed in the logistic regression method at Level 1, with a precision of 98.74%, a recall of 91.27%, an accuracy of 95.13%, and outstanding ROC-AUC and AUC-PR scores of 99.17% and 99.27%, respectively. These results demonstrate a significant enhancement over existing models, showcasing a superior balance in accuracy and computational efficiency and surpassing traditional single-model approaches.