Material property prediction of perovskite oxides based on machine learning

Perovskite oxides show great potential for applications in energy conversion and environmental protection due to their excellent catalytic properties and tunability. However, the process of screening stable perovskite oxides using traditional experimental and computational methods is time-consuming and laborious, limiting the development of their applications. This paper proposes a machine learning-based method for predicting the properties of perovskite oxide materials. Four machine learning models, random forest regression, gradient boosted regression, ridge regression, and support vector regression, were constructed using a dataset of ABO₃ perovskite compounds calculated by DFT by Antoine A. Emery and others, and the unit cell volume and tolerance factor were predicted. The results show that the random forest regression model achieved the best performance in predicting the unit cell volume and tolerance factor, with R² reaching 0.99932 and 0.99849, respectively, and MAE reaching 0.29832 and 0.0 0262. The model is explained based on the SHAP method, and it is found that the tolerance factor of perovskite oxides with A-site ion radii more significant than 1Å and B-site ion radii less than 0.8 Å is usually greater than 0.75, indicating that their crystal structures are relatively stable. The machine learning-based method for predicting the properties of perovskite oxide materials proposed in this paper can quickly screen out perovskite oxides with stable structures, providing meaningful theoretical guidance for accelerating research on efficient perovskite catalysts.