Specific surface area (SSA) of perovskites with uncertainty estimation approach

In recent years, the rapid development of computer technology and the Internet has enabled faster and more convenient data acquisition. Using big data and machine learning (ML) to accelerate the development of new materials is crucial. This study provides insight into the application of ML in predicting the specific surface area (SSA) of ABO3-type perovskite. In this article, we initially introduce the basic concepts of ML, its general workflow in materials research, and the specific algorithms used. We further discuss the structural features of ABO3-type perovskites and their development trends in photocatalytic applications. The prediction of SSA is vital for the application and performance optimization of perovskites, and ML provides an efficient and accurate means of prediction. In the modeling process, we collected experimental data on 50 ABO3-type perovskites from published literature and feature-engineered them in various ways to obtain three different datasets. By screening the features through principal component analysis (PCA) and Genetic Algorithm (GA), we found that GA is superior to the Random Forest (RF) model. At the same time, PCA is more suitable for the Support Vector Regression (SVR) model. The test results showed that the RF model has a prediction accuracy of 0.832, while the SVR model has a prediction accuracy of 0.808, and both methods exhibited high prediction accuracy. In addition, the leave-one-out cross-validation results further confirmed the robustness of both models. Finally, we conducted an in-depth analysis of the importance of features. The results showed that the calcination temperature (CT) and calcination time (AH) are the most critical features, which are positively and negatively correlated with the SSA of perovskite oxides, respectively. This provides valuable reference information for the design and optimization of perovskite materials. Overall, this study successfully predicted the SSA of ABO3-type perovskites using an ML method, which provides a new and effective tool for researching and applying perovskite materials.