Mapping between density of states and energy band gap for bismuth-based semiconductors via machine learning prediction

Abstract

Bismuth-based semiconductors have many applications in energy and environmental fields. Considering that the energy band gap (Ebg) control engineering is crucial for the structure–activity relationship, this paper constructs the mapping relationship between density of states (DOS) and energy band gap based on artificial intelligence machine learning (ML) algorithm. Common single ML models, including linear regression, support vector machine, k-nearest neighbor and gaussian processes regression, as well as ensemble algorithms such as random forest regression and gradient boosting machine, are specifically invoked. The results show that the random forest model, which belongs to ensemble algorithm, is superior to the single algorithm for the prediction performance and stability of the test set. And after the random search and grid search combined hyperparameter tuning operation, the average root mean square error (RMSE) is reduced from 0.340 to 0.281. This paper provides a new idea for material selection and experimental design of bismuth-based semiconductors.