Groundwater Potential Assessment Using Ann-Rks Machine Learning Based Models in Part of Aule Community, Akure, Nigeria

This study adopted an integrated approach combining hydrogeological, geophysical, and machine learning techniques to evaluate groundwater potential in Aule Community, Akure, Ondo State, Southwestern Nigeria. Remote sensing data, particularly the ASTER Global Digital Elevation Model (DEM), were processed in ArcMap 10.8.2 to derive key terrain conditioning factors such as slope and aspect, which influence surface runoff, infiltration, and groundwater recharge. The geophysical investigation involved electrical resistivity surveys using the Schlumberger array configuration, with 105 Vertical Electrical Sounding (VES) stations acquired across the study area. Interpretation of the VES data delineated five subsurface geoelectric layers: topsoil, lateritic layer, weathered layer, partially weathered or fractured basement, and fresh basement bedrock, providing detailed insight into aquifer geometry and subsurface conditions. These geoelectric and hydrogeological parameters were integrated as predictor variables within a supervised machine learning framework. A dataset comprising 40 matched well–VES control points were used for model training and validation, with hand-dug well measurements serving as the target variable. Two predictive models were developed: a conventional Artificial Neural Network (ANN) and an enhanced ANN integrated with Random Kitchen Sinks (ANN-RKS) to improve nonlinear feature representation. Using 5-fold cross-validation, the ANN-RKS model outperformed the standard ANN, achieving MAE = 0.056, MSE = 0.006, RMSE = 0.078, and R² = 0.910, compared to MAE = 0.082, MSE = 0.012, RMSE = 0.109, and R² = 0.801 for the ANN model. The optimized ANN-RKS outputs were subsequently spatially interpolated within a GIS environment to generate a detailed groundwater potential map of the study area.