Machine learning incorporating stability features and Bayesian Optimization for perovskite structure prediction

Abstract

In recent years, machine learning has been extensively applied in materials science research due to its outstanding capabilities in data processing and pattern recognition. This study aims to address the classification problem of perovskite crystal structures by introducing BO (Bayesian Optimization) for parameter tuning of CatBoost (Categorical Boosting) for the first time. Additionally, the stability of perovskites is innovatively incorporated as one of the key features alongside traditional features such as Electronegativity and Bond Length. This approach enables precise classification of perovskite structures into cubic, tetragonal, orthorhombic, and rhombohedral phases. Data standardization is performed using Robust Scaling, the class imbalance in the dataset was addressed using the ADASYN ( Adaptive Synthetic Sampling) during feature selection and the class_weight of the CatBoost during model training.Feature selection is conducted using RFECV (Recursive Feature Elimination with Cross-Validation). A comparative analysis of models based on the processed dataset demonstrates that the BO_CatBoost (Bayesian Optimized CatBoost) model, which includes the stability feature, achieves a classification accuracy of up to 86.89%, significantly outperforming traditional machine learning models.