Article’s

Customer churn remains one of the most significant threats to profitability in subscription-driven industries such as telecommunications. This paper presents an end-to-end analytics framework that combines descriptive business-intelligence dashboards with predictive machine learning and prescriptive decision support to address churn on a real-world telecommunications dataset comprising 7,043 customers and 21 attributes. Interactive dashboards were first developed to characterise churn across demographic, tenure, and service-contract dimensions, revealing an overall churn rate of 26.54%. To predict individual churn probability, three ensemble classifiers — Random Forest, XGBoost, and LightGBM — were trained on a class-balanced dataset produced using the Synthetic Minority Over-Sampling Technique (SMOTE), achieving a peak ROC-AUC of 83.5%. SHapley Additive exPlanations (SHAP) were applied to the best-performing model to render predictions interpretable, identifying contract type, monthly charges, and tenure as the dominant churn drivers. Building on these predictions, a rule-based prescriptive analytics engine automatically generated individualised retention offers for 1,585 high-risk customers, and a Customer Lifetime Value (CLTV) and risk-cohort segmentation prioritised 404 customers for immediate intervention. The results demonstrate that integrating descriptive, predictive, and prescriptive analytics into a single pipeline yields actionable, quantified retention strategies that outperform descriptive reporting alone.