Journal of Geography, Environment and Earth Science International

This study evaluates the effectiveness of a machine learning–based Gradient Boosting Tree (GBT) model for predicting groundwater potential and compares its performance with the conventional Analytical Hierarchy Process (AHP) model. Both approaches were applied using twelve groundwater potential predictors (GPPs): Digital Elevation Model (DEM), slope (S), drainage density (Dd), land use (Lu), aquifer resistivity (ρa), aquifer thickness (h), overburden thickness (b), hydraulic conductivity (k), transmissivity (Tr), storativity (St), diffusivity (D), and reflection coefficient (Rc). The GBT model was developed with Salford Predictive Modeler 8.0 using a 90:10 training–testing data split and k-10 cross-validation. Predictor importance was assessed, and the groundwater potentiality prediction index (GPPI) was generated and spatially analyzed in ArcGIS. Groundwater potential maps from both models delineated three zones: low, moderate, and high. The GBT-derived map indicated that low–moderate potential areas covered 56% of the study region, compared to 71% in the AHP-based map. Model validation using water column measurements from fifteen wells, analyzed with Spearman’s correlation, showed higher predictive accuracy for the GBT model (rs = 0.74; p = .002) than for the AHP model (rs = 0.66; p = .007). The results demonstrate that aquifer resistivity alone is inadequate for groundwater evaluation and underscore the importance of integrating multiple parameters. Overall, the GBT model outperformed AHP, establishing itself as a more reliable approach for groundwater potential prediction.