Rapid urbanization and industrial growth have intensified the demand for groundwater in Keraniganj, Bangladesh. This study integrates Geographic Information Systems (GIS), remote sensing (RS), the Analytic Hierarchy Process (AHP), and machine learning (ML) to delineate groundwater potential zones (GWPZ). Ten thematic factors, including geology, soil texture, rainfall, geomorphology, NDVI, TWI, lineament density, drainage density, land use/land cover, and slope, were weighted using AHP. Four ML models (Gradient Boosting Machine, Random Forest, Logistic Regression, and Decision Tree) were applied alongside GIS–AHP to evaluate predictive accuracy. The GIS–AHP model achieved 82.7% accuracy (Kappa = 0.77), confirming reliability but highlighting sensitivity to subjective weight assignments. The sensitivity analysis confirmed robustness of the AHP model, with less than 5% variation observed in high and very high groundwater potential zones, indicating stable confidence in the results. Among ML models, Gradient Boosting performed best (accuracy = 95%, ROC–AUC = 0.954), followed by Random Forest (94.8%, ROC–AUC = 0.937). ROC supported confidence in the machine learning outputs–AUC values (> 0.93) and tenfold cross‑validation, which consistently demonstrated high predictive reliability across different data subsets. Spatial analysis revealed high (22.9%) and medium (21.5%) potential zones concentrated in central and northwestern regions, while very high zones (17.3%) were localized. Low (27.0%) and very low (11.4%) zones dominated industrialized southern and eastern areas. Decision Tree performed weakest (AUC = 0.77). Beyond technical mapping, the study proposes four area-specific strategies for sustainable groundwater management, offering actionable guidance for policymakers and resource managers. By combining structured decision-making with adaptive ML, this research demonstrates a robust, cost-effective framework for groundwater assessment in rapidly urbanizing environments.
Safari et al. (Mon,) studied this question.