An Intelligent Evolving Ensemble Machine Learn-ing Framework for Customer Churn Prediction in Telecommunications
Authors-V.Kranthi Kumar, N sankeerthana
Keyword-Customer Churn Prediction, Machine Learning, Ensemble Learning, Neural Networks, Random Forest, XGBoost, K-Nearest Neighbors, Weighted Average Ensemble, Telecommunications, Predictive Analytics.
Customer retention has become one of the most critical challenges faced by telecommunication service providers due to intense market competition and continuously changing customer behav-ior. Accurately identifying customers who are likely to discontinue their subscriptions enables organizations to implement proactive retention strategies and minimize revenue losses. Tradition-al customer churn prediction models frequently struggle to adapt to dynamic customer behavior, resulting in reduced predictive performance over time. This research proposes an intelligent machine learning framework based on an Evolving Ensemble Predictor (EEP) that combines multiple predictive algorithms, including Neural Networks (NN), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and K-Nearest Neighbors (KNN), using a weighted average ensemble strategy. The proposed framework treats customer churn prediction as a continuously evolving problem rather than a static classification task. Historical customer records obtained from the Orange Telecom Churn Dataset are preprocessed through feature encoding, normaliza-tion, and feature selection before model development. Individual classifiers are trained inde-pendently, and their predictions are integrated using weighted ensemble learning to generate the final churn prediction. Experimental evaluation demonstrates that the proposed EEP model achieves improved prediction accuracy, precision, recall, F1-score, and computational efficiency compared with individual machine learning models. The proposed framework provides a scala-ble and adaptive solution for customer churn prediction, assisting telecommunication companies in improving customer retention, reducing operational losses, and supporting intelligent business decision-making while preserving the same implementation methodology and evaluation strategy as the original study.
Doi-[https://doi.org/10.5281/zenodo.21735322]