Studies on the regularity of perovskite formation via machine learning

Abstract

Perovskite material attracts great interest in many sciences and engineering fields. It is meaningful to improve the accuracy of judging whether ABX3 and A2B'B''X6 compounds can form perovskite structures. In this work, machine learning method, random forest classification (RFC) model was constructed to discriminate the formability of perovskites for both ABX3 and A2B'B''X6 compounds. Based on the six selected features, the accuracy of the RFC model is as high as 96.55% for ABX3 independent test set. Furthermore, RFC model can be successfully used to predict the formation of the perovskite structure of A2B'B''X6 compound with the accuracy 91.83%. Besides, by using RFC model, 241 ABX3 perovskites with a 95% probability of formability were screened out from 15,999 candidate compounds, meanwhile 1131 A2B'B''X6 perovskites with a 99% probability of formation were screened out from 417,835 candidate compounds. The method presented in this work could provide precious enlightenment for the acceleration of discovering the perovskite material.