Geoelectric curve amplitude and percentage thickening for aquifer vulnerability and hydrological connectivity in aquifer-water channel systems: Machine learning performance metrics in Akwa Ibom State

This study examines the geoelectric properties of aquifer systems in Akwa Ibom State, Nigeria, to assess vulnerability and hydrological connectivity in a coastal environment challenged by saltwater intrusion and fluctuating recharge–discharge dynamics. The objective is to determine how resistivity inversion, subsurface variability, and geoelectric indices can be applied to quantify aquifer protection and exposure to surface water influence. vertical electrical sounding (VES) and electrical resistivity tomography (ERT) were combined with the VES attendant curve amplitude and percentage thickening analyses to characterize aquifer depth, thickness, and resistivity across different layers. Aquifer depths ranged from 0.6 to 104.3 m, while resistivity values in unconfined freshwater aquifers varied between 35.4 and 104.3 Ωm. Coastal areas such as Oron, Eket, and Ikot Abasi showed pronounced saltwater intrusion, reducing both resistivity and usable aquifer thickness. Recharge inflow accounted for 56% against 44% discharge outflow, highlighting the dynamic interaction between rivers and aquifers in maintaining hydrological balance, while additional parameters including formation factor, porosity, hydraulic conductivity, and transmissivity further emphasized the spatial variability of groundwater flow and extraction potential. A tuned extreme gradient boosting (XGBoost) model was applied to assess aquifer vulnerability, and the receiver operating characteristic (ROC) curve produced a perfect training AUC of 1.00 and a robust test AUC of 0.875, demonstrating high classification accuracy and generalization. The confusion matrix recorded six correct predictions out of eight (accuracy = 0.75), with precision, recall, specificity, and F1-score each attaining 0.75, thereby confirming consistent identification of vulnerable aquifers with minimal misclassification of protected ones. Hydrogeologically, high amplitude values corresponded to protective clay overburden, whereas reduced percentage thickening indicated weaker barriers, greater connectivity with surface channels, and increased contamination risk. The innovation of this work lies in the integration of geoelectric curve amplitude and percentage thickening indices with machine learning performance evaluation, and by validating these indices through statistical performance metrics rather than qualitative interpretation alone, the study establishes them as reliable proxies for aquifer vulnerability and hydrological connectivity. The findings provide a scientific foundation for groundwater protection, sustainable abstraction, and long-term aquifer management in hydrologically sensitive coastal regions.