Ensuring clean water remains essential for public health and sustainable resource governance. This study presents an interpretable ensemble learning framework for predictive water quality assessment, integrating a hybrid feature selection approach, dynamic thresholding, and explainable AI techniques. By leveraging historical datasets and key physicochemical indicators, the proposed system employs models such as XGBoost, CatBoost, and ExtraTrees to improve classification accuracy and robustness. The use of SHAP enhances transparency by identifying critical features influencing potability predictions. Comparative evaluation against traditional models demonstrates the ensemble framework’s superior performance in accuracy, F1-score, and adaptability across varied environmental conditions. This work contributes a scalable, data-driven solution for proactive and sustainable water resource management.