An Interpretable Uplift Modeling Framework for Customer Churn Intervention Decision-Making

Customer churn challenges subscription-based industries, especially telecommunications. Traditional churn prediction identifies high-risk customers but cannot determine who will respond to retention campaigns. This paper proposes an interpretable decision support framework integrating predictive modeling, SHapley Additive exPlanations (SHAP) interpretation, and causal uplift analysis. An XGBoost classifier predicts churn probability on the public Telco dataset, while SHAP reveals tenure, contract type, and monthly charges as key drivers with nonlinear and interaction effects. Four uplift models—S-Learner, T-Learner, X-Learner, and Uplift Random Forest—are compared on a held-out test set using the Area Under the Uplift Curve (AUUC). The X-Learner performs best, achieving an AUUC of 0.724. A cost-sensitive simulation shows that targeting customers with high uplift scores improves expected profit substantially over treating all at-risk customers. The framework is implemented in a Streamlit dashboard, offering managers an actionable tool for churn prevention and shifting churn management from reactive prediction to proactive decision-making.