Enhancing Customer Churn Prediction Across Industries: A Comparative Study of Ensemble Stacking and Traditional Classifiers

Abstract

Predicting customer churn is essential in sectors such as banking, telecommunications, and retail, where retaining existing customers is more cost-effective than acquiring new ones. This paper proposes an enhanced ensemble stacking methodology to improve the prediction performance of ensemble methods. Classic ensemble classifiers and individual models are undergoing enhancements to enhance their sector-wide generalisation. The proposed ensemble stacking method is compared with well-known ensemble classifiers, including Random Forest, Gradient Boosting Machines (GBMs), AdaBoost, and CatBoost, alongside single classifiers such as Logistic Regression (LR), Decision Trees (DT), Naive Bayes (NB), Support Vector Machines (SVM), and MultiLayer Perceptron. Performance evaluation employs accuracy, precision, recall, and AUC-ROC metrics, utilising datasets from telecom, retail, and banking sectors. This study highlights the importance of investigating ensemble stacking within these three business entities, given that each sector presents distinct challenges and data patterns related to customer churn prediction. According to the results, when compared to other ensemble approaches and single classifiers, the ensemble stacking method achieves better generality and accuracy. The stacking method uses a meta-learner in conjunction with numerous base classifiers to improve model performance and make it adaptable to new domains. This study proves that the ensemble stacking method can accurately anticipate customer turnover and can be used in different industries. It gives firms a great way to keep their clients.