Crystal structural prediction of perovskite materials using machine learning: A comparative study. (15th February 2023)

• Record Type: Journal Article
• Authors: Priyadarshini, Rojalina; Joardar, Hillol; Bisoy, Sukant Kishoro; Badapanda, Tanmaya
• Is Part Of: Solid state communications. Volume 361(2023)
• Publication Date: 2023-02-15
• Subjects: Perovskite -- Machine learning -- Boosting algorithms -- K-nearest neighborhood -- XGBoost; Solid state chemistry -- Periodicals; Solid state physics -- Periodicals; Chimie de l'état solide -- Périodiques; Physique de l'état solide -- Périodiques; 530.41
• DOI: 10.1016/j.ssc.2022.115062
• Languages: English
• ISSNs: 0038-1098

Abstract

Abstract: In this study, Machine Learning (ML) techniques have been exploited to classify the crystal structure of ABO3 perovskite compounds. In the present work, seven different ML algorithms are applied to the experimentally determined crystal structure data. The relevance of the data featured is measured by computing the Chi-Square test and Spearman's correlation matrix. The Z-Score value has been calculated for each attribute to confirm the existence of any outliers in the data. The Synthetic Minority Oversampling (SMOTE) technique is employed to overcome the imbalanced data set. The models' performance is calculated using the stratified k-Fold cross-validation method. Further, to improve the accuracy of the prediction model, the conventional algorithm is supported by boosting algorithm. Comparative model efficiency on prediction of the crystal structure is presented to identify the most suitable model. As per the inferences drawn from the observations, the ensemble model using Xtreme Gradient Boosting (XGBoost) algorithm when applied to the pre-processed and balanced data outperforms the other models. Highlights: The prediction of crystal structure of ABO3 perovskite are implemented using various Machine Learning (ML) models. The relevance of the features is measured by computing the Chi-Square test and Spearman's correlation matrix. The SMOTE augmentation technique is employed to overcome the imbalanced of data set. The models' performance is accessed using stratified k-Fold cross validation method. The parameters obtained from various model are compared and XGBoost models is found to the most stable and accurate model.