Paper Title
Ensemble-XAI Churn Intelligence: A Profit-Centric Adaptive Framework for Telecom Customer Retention Using Dual-Layer Explainability and Real-Time Concept Drift Handling

Abstract
Customer churn remains the foremost revenue-risk challenge in the saturated global telecommunications market, where subscriber acquisition costs exceed retention expenditure by a factor of five to twenty-five [1]. Existing churn models prioritise raw accuracy while neglecting operational transparency and profit optimisation. This paper presents Ensemble-XAI Churn Intelligence, a unified framework that integrates a soft-voting ensemble (Random Forest + Gradient Boosting) with a Dual-Layer Explainable AI (XAI) module — SHAP for global feature attribution and LIME for instance-level explanation — and a novel Profit-Centric Adaptive Feedback Loop (P-AFL) that optimises interventions against Expected Maximum Profit for Churn (EMPC) [2]. Deployed as a FastAPI + React.js web platform, the system achieves recall above 75% on the churn minority class, delivers agent-readable per-customer explanations, and demonstrates a 63% improvement in EMPC per intervention after three adaptive retraining cycles. Automated concept drift detection via Population Stability Index (PSI) ensures the model remains calibrated under evolving customer behavioural patterns [3]. Keywords - Customer Churn, Explainable AI, SHAP, LIME, Ensemble Learning, Gradient Boosting, Random Forest, FastAPI, EMPC, Concept Drift, P-AFL, Telecom Retention