Efficient Prediction of Polymer Glass Transition Temperatures through Machine Learning Methods

Abstract

The glass transition temperature (Tg) plays a crucial role in defining polymer properties. Despite the widespread use of machine learning for material design and property prediction, there are still challenges concerning the interpretability and model performance when predicting Tg. In this study, Simplified Molecular Input Line Entry System strings are utilised to encode the polymer structure, which are then transformed into molecular descriptors for analytical training and prediction of Tg using Artificial Neural Network and Random Forest models. Meticulous hyperparameter tuning of the Random Forest model was performed, resulting in reasonable Tg predictions. This methodology forges a connection between polymer structure and Tg, opening up new avenues for research in the field of polymers.