Bandgap prediction of non-metallic crystals through machine learning approach.

Abstract

The determination of bandgap is the heart of electronic structure of any material and is a crucial factor for thermoelectric performance of it. Due to large amount to data (features) that are related to bandgap are now a days available, it is possible to make use of machine learning (ML) approach to predict the bandgap of the material. The study commences by selecting the feature through Pearson correlation study between bandgap and various thermoelectric parameters in non-metallic crystals. Among the 42 parameters available in the dataset, the Seebeck coefficient and its corresponding temperatures show high correlation with the bandgap. With these three selected features we have used different ML models like multilinear regression, polynomial regression, random forest regression and support vector regression to predict the bandgap. Amongst the different ML models considered, random forest regression outperforms the other models to predict the bandgap withR2value of 97.55% between actual bandgap and predicted bandgap.