Мachine learning methods for predict electrophysical properties of semiconductor materials for optoelectronic and energy storage devices

Abstract

There were several notable attempts at utilizing Machine Learning to predict physical properties of various materials. Huang et al. reported prediction of band gap properties for ternary metal nitride compounds using ML approach based on the calculated data using Heyd–Scuseria–Ernzerhof (HSE) hybrid functionals and Perdew–Burke-Ernzerhof (PBE) DFT methods. In that study electronegativity, valence and covalent radius were used as feature for the training of the ML algorithm and prediction. In another study, high accuracy of the prediction was achieved for the ML algorithm trained on the dataset with 3 only features such as ionic radius, electronegativity and number of row associated with position of specific element in the periodic table.