Machine learning approach for prediction of the grafting yield in radiation-induced graft polymerization

•Grafting yields of methacrylate ester monomers were predicted by machine learning.•Linear regression, decision tree, random forest and XGBoost algorithms were compared.•XGBoost and random forest models showed higher prediction accuracies.•Polarizability and O2 NMR shift were important for predicting grafting yield. Grafting yields for the radiation-induced graft polymerization of a methacrylate ester monomer to give a polyethylene-coated polypropylene nonwoven fabric were predicted as an objective variable by a machine learning approach. The degrees of grafting were obtained from actual experiments. Monomer structure information, atomic charge information, atomic NMR shift information, and infrared absorption wavenumber information, derived from density functional theory calculations, were adopted as explanatory variables of a grafting yield prediction model. Among machine learning algorithms as a prediction model on the grafting yield, XGBoost and random forest models showed higher prediction accuracy, compared to a multiple linear regression model. The prediction accuracies of the various algorithm decreased in the order: XGBoost > random forest > multiple linear regression/LASSO > decision tree > multiple linear regression. The monomer polarizability and the O2 NMR shift were found to be important explanatory variables for predicting the grafting yield in the XGBoost model. This is probably because the polarizability, which represents a miscibility indicator of the monomer to the trunk polymer, and the O2 NMR shift, which represents a diffusivity indicator of the monomer into the trunk polymer, remarkably reflect the difference in the substituent structure of the methacrylate ester monomers. [Display omitted]