A novel hybrid machine learning model for predicting rate constants of the reactions between alkane and CH3 radical

The rate constants of H-abstraction reactions of alkanes by free radicals are crucial for optimizing combustion reaction network, improving combustion efficiency and designing aero engines. In this study, three different machine learning (ML) algorithms-Feedforward Neural Network (FNN), eXtreme Gradient Boosting (XGB) and Random Forest -were used to develop ML models for predicting the rate constants of reactions of alkane and CH3 radical. The results showed that the FNN algorithm with 6 descriptors has better overall performance than the other two ML algorithms. A novel hybrid ML model combining FNN with XGB (XGB-FNN) was then developed, by which the prediction accuracy was visibly improved (p-value < 0.05). The average deviation of XGB-FNN model on the prediction set is 42.35%. The rate constants of the primary hydrogen abstraction reactions between CH3 radical and normal alkanes with 5 ∼ 12 carbon atoms were predicted over a temperature range of 300 ∼ 2500 K, which follow well the modified three-parameter Arrhenius equation and/or agree well with available experimental rate constants, indicating that the hybrid XGB-FNN model is robust in predicting the rate constants in the temperature range of combustion.