Predictive Analytics Framework for Strategic Decision-Making in Retail and Customer Churn Management

Customer relationship management continues to pose a substantial obstacle in the prediction of client attrition, particularly in sectors like telecommunications and e-commerce. This study introduces a hybrid prediction system that is reliable and incorporates XGBoost and MLP models using a soft-voting ensemble approach. When assessed on benchmark datasets, the proposed model achieved exceptional results, surpassing its competitors in terms of metrics such as 98.26% accuracy, 98.31% precision, 98.04% recall, 98.17% F1-score, and 0.9924 AUC. Comparing the ensemble method to previous models reveals that it strikes a satisfactory balance between interpretability and non-linear feature representation. This model is an excellent choice for enterprise-level attrition prediction systems due to its SHAP-based interpretability, real-time inference capacity, and tolerance to noisy data. This method is both adaptable and scalable, offering practical insights and predictive potential in a variety of customer-centric domains.