Can Machine\nLearning Algorithms Enhance the Prediction\nAccuracy of Linear Solvation Energy Relationship Models for Polyfluoroalkyl\nSubstances Adsorption by Activated Carbons in Complex Water Matrices?

Linear solvation energy relationship\n(LSER) models have\ntraditionally\nbeen used to predict the adsorption of organic contaminants (OCs)\non carbon-based adsorbents in pure water. However, predicting OC uptake\non solids is strongly influenced by the chemistry of water, adsorbent\ncharacteristics, and operational conditions. Machine learning (ML)-assisted\nLSER models can be promising solutions as an efficient tool to investigate\nthe fate and control of per- and polyfluoroalkyl substances (PFAS)\nin complex environmental settings. In this study, ML-assisted LSER\nmodels were investigated for the first time to predict PFAS adsorption\non activated carbons in complex water matrices. The results showed\nthat ML-assisted LSER models outperformed traditional LSER models,\nwith improved prediction accuracy (<i>R</i><sup>2</sup> =\n0.13–0.80 vs <i>R</i><sup>2</sup> < 0.1). Principal\ncomponent regression (PCR) was later applied to further enhance the\nefficiency of the ML models, resulting in more robust and accurate\npredictions (<i>R</i><sup>2</sup> = 0.65–0.99) through\na strategic combination of ML techniques. These combined approaches\nprovide valuable tools for investigating and controlling PFAS in environmental\ncompartments, providing new insights into developing source-tracking\nstrategies for managing PFAS.