Machine Learning Models for Tracking Pollutant (E. coli) Levels at the Water Intake at Lake Rådasjön

Ensuring safe drinking water requires accurate prediction of faecal contamination levels in water sources. Lake Rådasjön, a critical drinking-water reservoir in western Sweden, is subject to faecal contaminationfrom multiple sources, including on-site sewage systems, stormwater runoff, and inflows from the Mölndalsån River. Traditional hydrodynamic models have been employed in previous work to simulate thefate and transport of contaminants; however, they often require extensive parameterisation and computational resources.This study aims to develop and evaluate several machine learning (ML) models, Random Forest,Support Vector Machine, and XGBoost, to predict Escherichia coli (E. coli) concentrations at the waterintake of Lake Rådasjön. Historical monitoring data were used to train and validate the ML models.Feature-importance analysis identified key environmental drivers influencing E. coli transport, including wind speed, water temperature, and river inflows.Among the models tested, the binary XGBoost classifier showed the strongest ability to detect high E. coli concentrations. Although the models demonstrated promising predictive capabilities, their performance was limited by sparse microbial sampling and the complexity of lake dynamics.