Integrating Hydrochemical Characterization, Pollution Indices, and Machine Learning for Groundwater Quality Assessment in the Maadher Hodna Aquifer (Algeria)

Groundwater in the Maadher Hodna aquifer (Algeria) was assessed using an integrated approach combining hydrochemical analysis, pollution indices, GIS mapping, and machine-learning techniques to distinguish natural hydrochemical controls from nitrate-related anthropogenic impacts. Electrical conductivity exceeded the drinking-water guideline in 93.33% of samples, while nitrate concentrations exceeded 50 mg L⁻1 in 78.3% of samples. Groundwater was predominantly of the Ca–Mg–SO4–Cl type, reflecting carbonate and evaporite dissolution, with evaporation contributing to salinization in some wells. The Pollution Index of Groundwater (PIG) indicated mainly low to moderate overall groundwater pollution, whereas the Nitrate Pollution Index (NPI) classified 70% of wells as nitrate-polluted, indicating substantial localized anthropogenic influence. Support vector regression (SVR) showed weak and unstable nitrate prediction from routine hydrochemical variables under five-fold cross-validation ( $${R}^{2}=0.23\pm 0.30$$ ; two folds yielded negative $${R}^{2}$$ values), suggesting that nitrate variability is not adequately explained by background hydrochemistry alone. The support vector machine (SVM) efficiently reproduced PIG-based groundwater-quality classes from the hydrochemical variables used in the PIG calculation, demonstrating potential for automated screening rather than an independent predictive relationship. Overall, the integrated framework identifies widespread mineralization and spatially heterogeneous nitrate contamination and provides a basis for targeted groundwater monitoring and agricultural pollution management.