Enhancing the Predictive Kinetics of Intramolecular H-Migration Reactions of Ether Peroxy Radicals by Integrating Machine Learning with Quantum Chemistry: A Comparative Study of Generic Rate Rules and Machine Learning Techniques

Accurate reaction rate constants are essential for automated mechanism generators, yet high-level quantum chemistry methods remain computationally prohibitive. This study addresses this challenge for intramolecular H-migration in ether peroxy radicals, which are important in low-temperature combustion mechanisms. We investigated 102 reactions using M06-2X/6-311++G(2df,2p) calculations and canonical transition state theory. While we developed generic rate rules for 42 reaction classes, they proved inadequate for accurate predictions. Therefore, we built machine learning models with reaction-specific descriptors to predict the rate constants. We compared tree-based algorithms (eXtreme Gradient Boosting and Light Gradient Boosting Machine) against nontree-based algorithms (Multivariate Linear Regression and Gaussian Process Regression). The Light Gradient Boosting Machine emerged as superior with a root mean square error of 0.396 in ln k values. Feature importance and SHapley Additive exPlanation analysis revealed that structural characteristics and energetic properties were the most influential factors affecting reaction rates. Our systematic comparison demonstrates that machine learning methods significantly outperform generic rate rules for kinetic modeling of ether peroxy radical H-migration reactions.