Exploring the Classification of Crystal Structure of Perovskite Oxide Using Machine Learning and Deep Learning

Classification of crystal structures of perovskite oxides (ABO
3
) has been done using various machine learning algorithms. This paper identifies various features, like valance, ionic radius, and electronegativity from the dataset, and the respective features crucial for classification of the perovskites have been used to classify structures of perovskite oxides. We classified the oxides into cubic, tetragonal, orthorhombic, and rhombohedral crystal structures using machine learning and deep learning models. Amongst these, the Kernel SVM model proved to be the best model showing an accuracy of 94%. The GridSerachCV method was implemented for hyperparameter tuning. KernelSVM uses kernel trick to handle nonlinear data and thus allows an SVM to find a better hyperplane to classify. Therefore, KernelSVM proves to be a robust and reliable model for the classification of perovskite materials. Further, data augmentation improved the accuracy of all models except Logistic Regression. We also proposed a novel Deep Learning approach, Encoder-Decoder Kernel SVM (EDKSVM) which expands the feature matrix to an expanded dimension, outperforming the existing Machine Learning approaches.