Machine learning-driven prediction of heavy metal pollution for aquatic health surveillance and disease control

Heavy metal (HM) accumulation in lake sediments poses ecological risks and may increase aquatic organism vulnerability to disease, requiring cost-effective monitoring tools. This study developed an interpretable multi-output machine learning (ML) framework to predict sediment HMs (As, Cd, Cu, Hg, Ni, Pb, and Zn) in Swan Lake, using field variables (pH, dissolved oxygen, and temperature), sediment texture (median grain size), and aluminum content (Al%). Models with nested preprocessing and cross-validation achieved strong performance (R² = 0.824-0.997). Optimal algorithms varied by metal, while random forest showed the most consistent results, balancing accuracy, low error, and robustness to outliers. Predicted concentrations were integrated with ecological risk indices (Igeo, PLI, and Hakanson RI), indicating low pollution with localized Zn hotspots. Explainability analysis identified pH and Al% as key drivers, supporting ML-based screening for aquatic monitoring and early warning systems.