The prediction of heavy metal contamination in groundwater using machine learning algorithms: a case study from the Harran Plain, a major agricultural irrigation area in Türkiye.

This study aims to determine the current status of groundwater in terms of heavy metal pollution in Harran Plain, which has been subjected to agricultural irrigation for over thirty years and is exposed to point and diffuse pollutant pressure. In this context, groundwater samples were taken from 26 sampling points during the irrigation season and heavy metal parameters such as Ag, Al, B, Ba, Cd, Co, Cr, Cu, F, Fe, Li, Mn, Mo, Ni, Pb, Zn were analyzed. The pollution indices (Heavy metal pollution index (HPI), Heavy metal evaluation index (HEI) and Contamination index (Cd)) of the obtained data were calculated and Machine Learning (ML) models such as Random Forest (RF), Support Vektor Machines (SVM), K-Nearest Neighbors (KNN), Decision Tree (DT), and Logistic Regression (LR) classifiers were used to estimate these indices and the model performances were evaluated by comparing them. According to the national and international legislation evaluation, EC and TDS values were observed at high levels at sampling points G10 (Arican), G13 (Gulveren) and G19 (Bugdaytepe). It was determined that the permissible limit values were exceeded in the parameters Al, Fe, F, B, Zn, Pb, Mn and Mo. Accordingly, the heavy metal concentration order determined above the limit values was as follows: Fe > F > Al > B > Zn > Pb > Mo > Mn. The total metal load of the water samples was largely in the Near Neutral - High Metal class. According to the highest percentages for pollution indices, "Low" for HPI, "Medium" for HEI and "High" for Cd were determined. In machine learning algorithms; RF and DT showed the best performance in HPI and HEI estimation, while DT, RF and LR models were found to be quite effective in CD estimation. SHAP emphasizes that Mo is generally one of the most effective parameters for HPI, HEI and CD index evaluation in DT and RF classification. In order to ensure sustainable environmental health and agricultural development in the plain, it is recommended to increase monitoring programs especially for heavy metal and other pollutant parameters.