Groundwater Hydrogeochemistry, Uranium Mobility and Human Health Risk in the Mikir Hills, Assam, India: An Integrated Multivariate, Machine-Learning and Risk-Assessment Framework

Groundwater in the Mikir Hills of Assam, India, occurs within a complexly layered lithostratigraphic framework spanning Archaean basement gneisses to Quaternary alluvium.To characterise the hydrogeochemical controls on major ions, trace metals, and uranium (U) behaviour, a dataset of 143 groundwater samples across 16 core parameters (U, Na, K, Ca, Mg, Cl, SO₄²⁻ , HCO₃⁻ , pH, electrical conductivity, TDS, Cu, Pb, Zn, Ni, and Fe) was evaluated using an integrated multivariate, geochemical, and machine-learning framework.Hydrochemical facies analysis demonstrates that groundwaters are predominantly of Ca-Mg-HCO₃⁻ (52%) and mixed Ca-Na-HCO₃⁻ (30%) types, with Gibbs formulations confirming rock-weathering dominance alongside negative chloro-alkaline indices denoting reverse ion exchange.Speciation modeling indicates widespread supersaturation with respect to calcite (SI=+0.42)and dolomite (SI=+0.77),confirming that carbonate buffering establishes circumneutral-to-alkaline conditions.Nonparametric correlation and principal component analysis (PCA; four components explaining 67.2% of total variance) show that U is decoupled from bulk mineralisation (loading -0.04 on PC1) and instead loads on PC3 (0.64) and PC4 (0.35) alongside Cu, Fe, and Zn.Random Forest (RF) regression reveals negative cross-validated performance (R2=-0.077),whereas RF classification effectively discriminates upper-quartile U concentrations (>0.50 μg/L) with a ROC-AUC of 0.794.