Crystal structure classification in ABO3 perovskites via machine learning

Abstract

Crystal structure classification of perovskites (ABO3) is done using the Light Gradient Boosting Machine (Light GBM) algorithm. In this work, we have identified features such as electronegativity, ionic radius, valence, and bond lengths of A-O and B-O pairs that enable a priori crystal structure prediction. We have taken 5329 ABO3 perovskites and applied the proposed model to 675 compounds. It successfully categorized the compounds into cubic, tetragonal, orthorhombic, and rhombohedral structures with 80.3% best accuracy using 5-fold cross-validation. Therefore, the model can be used as a preliminary, fast, and inexpensive method to classify perovskites into their respective crystal systems. Feature importance graph and SHapley Additive exPlanations (SHAP) are used in feature ranking and crystal structure prediction. These composition-structure predictions will find applications in ceramic engineering and solid-state chemistry.