Can Machine Learning Algorithms Enhance the Prediction Accuracy of Linear Solvation Energy Relationship Models for Polyfluoroalkyl Substances Adsorption by Activated Carbons in Complex Water Matrices?

Linear solvation energy relationship (LSER) models have traditionally been used to predict the adsorption of organic contaminants (OCs) on carbon-based adsorbents in pure water. However, predicting OC uptake on solids is strongly influenced by the chemistry of water, adsorbent characteristics, and operational conditions. Machine learning (ML)-assisted LSER models can be promising solutions as an efficient tool to investigate the fate and control of per- and polyfluoroalkyl substances (PFAS) in complex environmental settings. In this study, ML-assisted LSER models were investigated for the first time to predict PFAS adsorption on activated carbons in complex water matrices. The results showed that ML-assisted LSER models outperformed traditional LSER models, with improved prediction accuracy (R2 = 0.13–0.80 vs R2 < 0.1). Principal component regression (PCR) was later applied to further enhance the efficiency of the ML models, resulting in more robust and accurate predictions (R2 = 0.65–0.99) through a strategic combination of ML techniques. These combined approaches provide valuable tools for investigating and controlling PFAS in environmental compartments, providing new insights into developing source-tracking strategies for managing PFAS.