Innovative Feature Engineering Driven by Chemical Category in Machine Learning for Optimizing the Prediction of Hydroxyl Radical Reaction Rate Constants

Hydroxyl radical (HO•), renowned for its high reactivity and broad-spectrum oxidation capacity, is pivotal in advanced oxidation processes for degrading organic pollutants. However, accurately predicting the reaction rate constants between organic matter and HO• remains a significant challenge. This study first presents a machine learning-based predictive framework that integrates molecular fingerprints with chemical category features through innovative feature engineering. By systematically optimizing feature selection and encoding, the framework enhances both predictive accuracy and generalizability. With an R2 value of 0.9794 and a root mean square error of 0.1313 for predicting the reaction rate constants between HO• and organic chemicals on the test set, the model demonstrates outstanding performance. Additionally, the framework captures reactivity patterns across diverse organic categories and elucidates the molecular and categorical factors driving HO• reactions. Interpretability analysis identifies key features influencing reactivity, offering insights into the behavior of organic pollutants and enhancing model transparency. This work establishes a robust methodological foundation for organic pollutant degradation kinetic modeling with accurate prediction of the reaction rate constants between HO• and organic matter, which benefits the advancements in water quality management and pollution control.