Machine Learning Models for Predicting Persistent Organic Pollutants in Water Bodies

Abstract

Persistent Organic Pollutants (POPs) are a class of toxic, bioaccumulative, and environmentally persistent contaminants that pose significant threats to aquatic ecosystems and public health. Traditional monitoring approaches for assessing POP concentrations in water bodies are often labor-intensive, time-consuming, and economically demanding, limiting their effectiveness for large-scale environmental surveillance. Recent advancements in machine learning (ML) provide promising alternatives for predicting contaminant behavior using complex environmental datasets. This study explores the application of machine learning models for predicting POP concentrations in water bodies by integrating physicochemical, hydrological, meteorological, and land-use variables. Various supervised learning algorithms, including Random Forest, Support Vector Machine, Gradient Boosting, Extreme Gradient Boosting, and Artificial Neural Networks, are evaluated for predictive performance. The proposed framework emphasizes data preprocessing, feature engineering, model optimization, and interpretability analysis to improve prediction accuracy and environmental decision-making. Comparative assessment demonstrates that ensemble learning approaches achieve superior predictive capability and robustness under varying environmental conditions. The findings highlight the potential of ML-driven predictive systems to support early warning mechanisms, pollution management strategies, and sustainable water resource governance. The study contributes toward the development of intelligent environmental monitoring systems capable of enhancing the detection and prediction of POP contamination in aquatic environments.