EXTREME GRADIENT BOOSTING (XGBOOST) ALGORITHM FOR CUSTOMER CHURN PREDICTION AND RESOURCE OPTIMIZATION IN TELECOMMUNICATIONS (A CASE STUDY OF MTN CONNECT STORE YOLA)

ABSTRACT This study developed and evaluated a predictive model for customer churn using the Extreme Gradient Boosting (XGBoost) algorithm to support resource optimization at MTN Connect Store, Yola. The research was motivated by the critical need for telecommunications companies to proactively manage customer retention in a highly competitive market. Utilizing a dataset of 100 customers with demographic and behavioral attributes, the study addressed the significant class imbalance between churners and non-churners by applying the Synthetic Minority Over-sampling Technique (SMOTE) prior to model training. The implemented XGBoost model demonstrated exceptional performance, achieving an overall accuracy of 94%, with a precision of 96% and a recall of 92% for identifying churners. Feature importance analysis revealed that Offer Response was the most influential predictor of churn, followed by Tenure in Months and Monthly Spending (NGN). These findings provide actionable insights, indicating that customer engagement with promotional offers and the spending patterns of newer subscribers are primary indicators of attrition risk. The study concludes that the integration of XGBoost with SMOTE provides a robust, interpretable, and effective framework for predicting customer churn at a localized branch level. It is recommended that MTN Connect Store, Yola, adopts this model to enable data-driven decision-making, allowing for the targeted allocation of retention resources to high-risk customers, thereby minimizing revenue loss and enhancing customer loyalty. This research serves as a replicable model for applying advanced machine learning to customer relationship management in the Nigerian telecommunications sector.