Impacts and sources of potential toxic elements on water quality and optimizing machine learning models for sustainable management

Abstract This study developed an integrated hydrogeochemical-machine learning approach for groundwater quality assessment and health risk evaluation in Yemen’s Al Jawf basin. Thirty-three samples from the Quaternary aquifer were analyzed for major ions and trace metals, with multivariate statistics identifying hydrogeochemical patterns and Random Forest (RF) models predicting water quality indices. Hydrogeochemical analysis revealed three water types: Ca-HCO₃ (45%), mixed Ca-Mg-HCO₃-SO₄ (36%), and Na-Cl (19%) facies, controlled by carbonate dissolution, silicate weathering, and anthropogenic inputs. Principal component analysis identified three components explaining 78.5% of variance, with natural mineralization dominating (45.2%). Irrigation water quality index (IWQI) assessment classified 52% of samples as excellent-good, 33% as moderate, and 15% as poor-very poor. Significant samples (63.5-93%) in eastern and northwestern areas showed potential negative impacts on soil fertility based on irrigation indices. Health risk evaluation revealed that 27% of samples exceeded drinking guidelines for Fe and Mn, with children more vulnerable than adults according to hazard index value (HI > 1). Random Forest models achieved high accuracy (R² = 0.94–0.97) predicting multiple indices using 6–8 parameters, reducing monitoring costs by 60%. This represents the first post-conflict comprehensive assessment in Al Jawf basin, providing a replicable framework for sustainable groundwater management in arid regions with limited resources.