Enhancing Customer Churn Prediction with Explainable AI, Interactive Dashboard Analytics, and Personalized Retention Strategies in ECommerce

Abstract

Customer churn in online retailing is therefore a complex issue going beyond the binary classification approach needing join frameworks integrating predictive precision, time-oriented forecasting, comprehensibility and business smarts that can be acted upon. Modern churn prediction systems fail to provide transparency in the explanation and are unable to provide a time horizon as well as being unable to figure out what individual companies need to do to retain customers. The proposed research is a standalone AI framework, which combines supervised machine learning algorithms, survival analysis (Kaplan-Meier and Cox Proportional Hazards models), and explainable AI (SHAP-based). Customer risk profiles generated as the outputs of a risk model, and SHAP attires are matched by a personalization engine with retention strategies. The proposed solution is implemented since the modular, real-time architecture with RESTful APIs and an interactive dashboard that is developed with the Stream lit framework and tested on real-life ecommerce data. XGBoost outperformed the baseline models, achieving an accuracy of 95.8%, with a precision of 0.849, recall of 0.916, and an F1-score of 0.881. It works well with complex data and can handle imbalanced classes effectively. The model identified important features that influence customer churn, helping businesses focus their retention efforts.