Customer churn poses a significant challenge for the telecommunications sector, resulting in substantial revenue losses and increased customer acquisition costs. This research creates an efficient churn prediction model that combines state-of-the-art machine learning with ensemble learning to maximize customer retention. With the IBM Telco Customer Churn dataset, several baseline models, including Gradient Boosting, AdaBoost, Logistic Regression, Random Forest, and Support Vector Classifier, were compared with a suggested ensemble model that integrates stacking and soft voting. A comparative analysis of AUC, Average Precision, Precision, Recall, and F1-score reveals that although boosting-based methods yield competitive results, the proposed ensemble model decisively surpasses all baselines, with an AUC of 92.06 and an F1-score of 86.45. By leveraging solutions such as class imbalance, feature redundancy, and model interpretability, the framework enables the gathering of actionable insights for early churn prediction and focused retention strategies. The results emphasise the value of ensemble learning in providing strong predictive accuracy and business value, aligning with the sustainable development principles of telecommunications.
Goyal et al. (2026) studied this question.