Predicting Groundwater Quality Using Machine Learning Models: A Case Study in Semnan Province, Iran

Groundwater in arid and semi-arid regions faces increasing pressures from overexploitation, climate variability, and declining quality, making accurate forecasting essential for sustainable management. This study examines two decades of hydrological and chemical data from Semnan Province, Iran (2003–2023), including precipitation, piezometric levels, and key water quality parameters. Eight nonlinear machine learning models, covering tree-based, ensemble, kernel-based, instance-based, and neural networks, were tested. Temporal patterns were captured using a walk-forward validation combined with exponentially weighted averaging. Ensemble methods, especially Random Forest and XGBoost, delivered the best performance, with an average R² of 0.90, RMSE of 820.0, and RMAE of 21. These results show that combining different environmental data with advanced nonlinear models provides a reliable and flexible way to forecast groundwater quality in water-scarce areas.