Machine-learning-based prediction of cubic perovskite formation energy and magnetism
Perovskite materials have garnered considerable attention due to their excellent properties.Thus,the accurate prediction of their physical properties is paramount.In this study,we developed an element attribute matrix descriptor via element attribute addition or phase division to improve the prediction efficiency of formation energy and magnetism of cubic perovskite.For this purpose,we trained three machine-learning models,namely extra tree,gradient boosting,and multi-layer perceptron,based on 18928 cubic perovskite data.Subsequent evaluation was conducted to evaluate the predictive efficiency of formation energy,volume-normalized magnetic moment,and magnetic classification.The results revealed that the efficiency of the developed descriptor was far higher than that of element attribute statistics and force-field inspired descriptors.The R2 fractions for the prediction of formation energy and magnetization were found to be 97％and 80％,respectively.In addition,the AUC of magnetic/non-magnetic classification reached 93％,with the accuracy exceeding 90％.Thus,this paper provides an important reference for the prediction of perovskite materials based on machine learning by constructing a simple and effective material descriptor.
万方数据
维普
