Advanced water quality assessment using machine learning: Source identification and probabilistic health risk analysis
Amin Mohammadpour, Ehsan Gharehchahi, Mohammad Golaki, Majid Amiri Gharaghani, F. F. Alireza Ahmadian, Soroush Abolfathi, Mohammad Reza Samaei, Md Galal Uddin, Agnieszka I. Olbert, Amin Mousavi Khaneghah
Abstract
• Extra Trees Algorithm provides optimal predictive efficacy for Water Quality Index. • The majority of samples exhibited elevated total dissolved solid concentrations. • The RMS-WQI classification primarily indicated a 'Fair' water quality status. • EC, Cl, and TDS were identified as the primary influencers of WQI. • Non-carcinogenic risk indicated significant health risks, predominantly for children. Water resources and their quality are paramount for urban development and maintaining ecological health, particularly in arid regions confronting water scarcity. This study assessed groundwater quality in water-stressed region in southern Iran using the newly developed Root Mean Square Water Quality Index (RMS-WQI) model in conjunction with a health risk assessment (HRA) to evaluate potential risks to human health. Analysis of groundwater samples revealed that approximately 99.41 % of sites met the permissible limits for pH, fluoride ( F − ), and nitrate (NO 3 − ). Total dissolved solids (TDS) exceeded the recommended guidelines at nearly 63.90 % of locations. The RMS-WQI classified groundwater quality as ranging from "marginal" to "good", with scores between 43.20 and 85.33 (averaging 62.91±9.33). The Extremely Randomized Trees (ExT) algorithm demonstrated strong predictive capability for RMS-WQI, with sensitivity analysis identifying electrical conductivity (EC) and chloride (Cl − ) as the most influential parameters. The HRA results indicated notable health risks from F⁻ and NO₃⁻ exposure, particularly among children, where the hazard index (HI) exceeded the safety threshold at 57.4 % of sites. Ingestion rate (IR) was the dominant contributor to HI across all age groups. NaCl is found to be a major constituent of the regional groundwater. These findings highlight the efficacy of integrating RMS-WQI with machine learning tools for a robust assessment of groundwater quality and associated health risks in arid environments.