The success of the Materials Genome Initiative [1] has led to opportunities for data-driven approaches for materials discovery. The recent development of Polymer Genome (PG), which is a machine learning (ML) based data-driven informatics platform for polymer property prediction, has significantly increased the efficiency of polymer design. Nevertheless, continuous expansion of the ‘training data’ is necessary to improve the robustness, versatility and accuracy of the ML predictions. In order to test the performance and transferability of the predictive models presently available in PG (which were previously trained on a dataset of 451 polymers), we have carefully collected additional experimental glass transition temperature (Tg) data for 871 polymers from multiple data sources. [2,3] The Tg values predicted by the present PG models for the polymers in the newly collected dataset were compared directly with the experimental Tg to estimate the accuracy of the present model. Using the full dataset of 1322 polymers, a new ML model for Tg was built following past work [4]. The RMSE of prediction for the extended dataset, when compared to the earlier one, decreased to 21.0 K from 48.7 K.To further improve the model’s performance, we are continuing to accumulate new data and exploring new ML approaches.

Assessing and improving machine learning model predictions of polymer glass transition temperatures

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.