PSO-SVR predicting for the Ehull of ABO3-type compounds to screen the thermodynamic stable perovskite candidates based on multi-scale descriptors

we explored the correlation between the stability and descriptors with ML model. The framework of ML prediction for Ehull and screening stable ABO3 candidates for perovskite as described in Fig. 1. [Display omitted] Most of the research on perovskite materials rely on costly experiments or complex density functional theory (DFT) calculations to a large extent. In contrast, machine learning (ML) combined with data mining is more effective in predicting perovskite properties. In this work, by mining data from the Materials Project database and other materials databases, we constructed a raw data set containing the ABO3-type compounds calculated by density functional theory (DFT) and generated a feature set based on multi-scale descriptors including compound properties and component element attributes. By comparing various machine learning models, the optimized support machine regression (SVR) model, Particle swarm optimization-support machine regression (PSO-SVR) were used to predict the energy above the convex hull (Ehull) of ABO3-type compounds that is the criteriafor thermodynamic stability of ABO3-type compounds. In addition, the important descriptors that have significant influence on the thermodynamic stability of ABO3-type compounds were screened out, and the relationship between these descriptors and Ehull was discussed. Finally, the stable and ideal ABO3 compounds were screened out for perovskite candidates.