Journal of Advanced Robotics, Autonomous Systems and Human-Machine Interaction
Domain-Adaptive Customer Churn Prediction with Integrated SHAP-Based Explain ability
Abstract
Vaibhav Hingnekar, Sarthak Ghadge and Yash Jadhav
Predicting when customers are about to leave is a crucial challenge for businesses in telecommunications, banking, e-commerce, and beyond. Although machine learning has vastly improved our ability to forecast churn, many existing solutions come with significant barriers—such as limited adaptability to new industries, lack of transparency, and complicated deployment. This paper introduces a streamlined, client-side web application designed for versatile, single- customer churn prediction—complete with easy-to-understand, SHAP-inspired explanations and actionable retention suggestions. Our system offers four domain-specific models (Telecom, Banking, E-commerce, and General), each equipped with curated features and feature importance weights. The user experience is simplified into three sequential steps: login, model selection, and prediction. With clear, transparent feature reasoning and no need for backend or database integration, this application bridges the gap between advanced analytics and real-world business usability. Illustrative examples and comparison with traditional approaches demonstrate how our system makes sophisticated churn prediction accessible, interpretable, and practical for actual business decision-makers.