Machine learning in ecotoxicology: Pollutant exposure levels and detection, biotoxicity and environmental behavior...

In recent years, the worsening problem of environmental pollution and the limitations of traditional toxicological assays have accelerated the adoption of machine learning (ML) in ecotoxicology. ML enables rapid and accurate prediction of pollutant exposure and toxicity through the analysis of large-scale heterogeneous datasets. It demonstrates significant advantages particularly in estimating exposure levels and biological toxicity of contaminants such as pesticide residues, antibiotics, and heavy metals. By constructing data-driven models, ML enhances toxicity prediction across single-, multi-, and cross-species contexts and supports extrapolation to untested organisms. ML is also increasingly applied to model the environmental behavior of pollutants under dynamic conditions. Recent advances emphasize the importance of standardized data preprocessing and species-aware partitioning to improve reproducibility, while interpretability methods such as SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) link predictive features to mechanistic toxicology. Incorporating species traits and multi-task learning strengthens ecological validity, and integration of environmental covariates (e.g., pH, temperature, dissolved oxygen) refines predictions of pollutant fate. Finally, rigorous validation using external test sets, baseline comparisons, and calibration metrics ensures robust and regulatory-relevant applications.