A Gradient Boosting-Based Machine Learning Model for the Prediction of the Hardness of Thermoplastic Polymers

Abstract

Modern experimental techniques can measure the hardness of thin films and structural features on the nanometer scale. These techniques can be applied to many types of materials, but as the hardness of the material decreases, more specialized and precise equipment is needed. For polymers, this analysis can be relatively expensive and time-consuming, so here we demonstrate a gradient-boosting-based machine-learning model that can provide a fast and reliable prediction of the hardness of thermoplastic polymers. This model is based on the XGBoost algorithm and shows reasonably good performance with an R2 value of 0.46–0.69. The features used in this study include experimental (mechanical and thermal properties) and molecular descriptors obtained using RDKit. The model is robust, with reasonable performance obtained when key experimental features are removed and/or when the selection of polymers is reduced. This suggests that this approach will be useful in “bottom-up” strategies for polymeric materials development.