Prediction of seasonal drinking groundwater quality index using explainable hybrid XGBoost with novel advanced metaheuristic optimization algorithms in a semi-arid region

In semi-arid regions, groundwater quality exhibits complex nonlinear behavior and strong seasonal variability. Accurate prediction is therefore essential for sustainable management. This study proposes an explainable hybrid framework that integrates XGBoost with three advanced metaheuristic optimization algorithms: the Honey Badger Algorithm (HBA), Marine Predators Algorithm (MBA), and Coati Optimization Algorithm (CAO). The framework predicts Drinking Groundwater Quality Index (DGWQI) using 104 samples from 52 monitoring wells in the Melur aquifer, during both the pre-monsoon (PRM) and post-monsoon (POM) seasons in a semi-arid region, Tamil Nadu, India. The optimized models were systematically evaluated for predictive accuracy, convergence behavior, computational efficiency, generalization capability, robustness, explainability, and spatial prediction reliability. Among the evaluated models, XGB-HBA achieved the best predictive performance for the PRM dataset (R 2 = 0.990, RMSE = 3.794, and MAE = 3.171). In contrast, XGB-MBA produced the best result for the POM dataset (R 2 = 0.9773, RMSE = 5.40, and MAE = 3.828). SHAP analysis revealed distinct seasonal control on groundwater quality. TDS was the dominant predictor during the PRM, whereas groundwater quality in the POM season was influenced by a broader combination of hydrogeochemical variables associated with recharge processes. Groundwater quality maps consistently identified both persistent hotspots and seasonal spatial variations.