A Comprehensive Risk Assessment Framework for Groundwater Quality: Integrating Natural Background Levels, Threshold Values, and Machine Learning-Based Hydrochemical Facies Classification.

Groundwater is a critical freshwater resource increasingly affected by anthropogenic activities. This study introduces an innovative and integrated framework for assessing the groundwater quality of the Qazvin aquifer, Iran, by combining iterative statistical methods and advanced machine learning (ML) algorithms. Three complementary statistical techniques, i.e., Iterative 2σ, cumulative distribution function (CDF), and iterative Grubbs test (IGT), were employed to determine natural background levels (NBLs) and threshold values (TVs) for key physicochemical parameters, supported by validation through ProUCL 5.2. The IGT method proved most effective in eliminating anthropogenic outliers and capturing the true geogenic baseline. Temporal and spatial variations of NBLs indicated the combined effects of geological, climatic, and human factors on groundwater chemistry. Hydrochemical facies were subsequently classified using four ML models (SVM- and CatBoost-based) integrated with a compositional log-ratio (7hlr) transformation. The SVM-based models achieved the highest classification accuracy (> 93%) and Kappa coefficients (> 0.85), outperforming classical graphical approaches. The key innovation of this study lies in the dual integration of iterative statistical tools and ML-driven classification, offering a more accurate distinction between geogenic and anthropogenic processes. This framework provides a robust and transferable approach for evaluating groundwater quality and supports evidence-based strategies for sustainable aquifer management.