Facilitating polymer property prediction with machine learning and group interaction modelling methods

Identification of a suitable polymer material for a given application requires information about the properties and behavior of the material, which is time-consuming and costly to measure experimentally. In this study, we explore two computational alternatives; namely, group interaction modelling (GIM) and Machine Learning (ML) approaches, as two avenues for predicting six different thermal and mechanical properties of polymers. Random Forest (RF) was employed as ML algorithm. Molecular descriptors for ML and physical input parameters for GIM method were obtained directly from the chemical structure information of polymers. The ML models developed in this study exhibited strong predictive performance, achieving R2 values ranging from 0.83 to 0.955 across the evaluated properties. The accuracy of the ML and GIM method has been compared with each other, and the evaluation is demonstrated that ML approach offers more reliable predictions over the GIM method. Further-more, we found that the accuracy of the GIM predictions was highly dependent on the accuracy of the Debye temperature values used as an input parameter, particularly for predicting the glass transition temperature. Therefore, for better prediction in GIM procedure, it is essential to use accurate techniques to find the Debye temperature values.

Automated summary

This study explores group interaction modelling (GIM) and machine learning (ML) approaches for predicting thermal and mechanical properties of polymers. ML approach offers more reliable predictions compared to GIM, which is highly dependent on the accuracy of input parameters.