Comparative Analysis of Machine Learning Models for Telecom Churn Prediction

This research paper presents a comparative analysis of machine learning models for telecom customer churn prediction, focusing on improving both predictive performance and business decision-making. Customer churn is a major challenge in the telecommunications industry as it directly impacts revenue, profitability, and customer retention strategies. The study evaluates three models—Logistic Regression, XGBoost, and Neural Networks—to identify customers who are likely to discontinue services based on customer demographics, billing details, service usage patterns, and contract information. The proposed framework extends traditional churn prediction by incorporating cost-sensitive threshold optimization, where business factors such as customer contact cost, retention value, and retention success rate are considered alongside predictive accuracy. Instead of relying only on metrics like accuracy, precision, recall, and ROC-AUC, the system evaluates expected profit at different decision thresholds to determine the most effective intervention strategy. Experimental results show that XGBoost achieves the highest expected profit while requiring fewer customer contacts compared to Logistic Regression and Neural Networks, demonstrating that threshold optimization can be as important as model selection in practical churn prediction systems.