Classification of magnetic ground States and prediction of magnetic moments for ABX3 perovskites utilizing machine learning techniques

Abstract

Perovskite magnetic materials have garnered significant attention due to their excellent applications in spintronics and magnetic storage. However, their discovery via conventional methods often entails considerable computational cost and time. This study employs machine learning (ML) to develop a classification model for magnetic ground states and a regression model for predicting the magnetic moments of perovskite materials, utilizing a dataset of 1511 ABX₃ compounds. We systematically evaluated four decision tree algorithms: Extra Trees (ET), Gradient Boosting (GB), eXtreme Gradient Boosting (XGB), and Random Forest (RF). The XGB classification (XGBC) model exhibited the best performance (accuracy = 0.946) for classifying ferromagnetic (FM), ferrimagnetic (FiM), antiferromagnetic (AFM), and non-magnetic (NM) states. While the XGB regression (XGBR) model demonstrated excellent performance (maximum R² = 0.919) in predicting magnetic moments. The XGBC model was subsequently applied to classify the magnetism of 304 new ABX₃ compounds, further validating its effectiveness as a screening tool for identifying perovskite materials with distinct magnetic properties. Finally, SHapley Additive exPlanations (SHAP) analysis underscored the significant role of perovskite’s magnetization in magnetism. This study highlights ML’s potential for analyzing ABX₃ perovskites magnetism and provides a valuable approach to accelerating the discovery and design of novel ABX₃ perovskite magnetic materials.