Comparative Analysis of Conventional Machine Learning and Graph Neural Network Models for Perovskite Property Prediction

Perovskite materials, renowned for their versatility and remarkable properties, pose challenges in discovering optimal candidates due to the vast compositional space. Data-driven machine learning (ML) offers promise in expediting material discovery; however, the trade-off between accuracy and efficiency across different ML models for predicting perovskite properties is not well-understood. In this study, we conducted a comprehensive assessment of various ML models for predicting the formation energy (Ef) and band gap (Eg) of perovskites. We designed a protocol to extract perovskite structures from three databases based on the stoichiometry, octahedral lattice motif, and alignment with established perovskite prototype structures. Benchmarking conventional ML algorithms (CML) against graph neural network (GNN) models across three data sets, we identified LGBM and GATGNN models as the top performers for CML and GNN, respectively, balancing exceptional prediction accuracy and computational efficiency. We further investigated the impact of the data size on model performance, emphasizing the need for over 1000 data points for optimal prediction accuracy. Additionally, through SHAP analysis, we provided valuable insights into the interpretation of CML models in predicting Ef and Eg. Our study establishes a standardized benchmark for evaluating various ML models across diverse data sets of perovskite materials, facilitating future applications in materials science, particularly in model selection and advancement of perovskite materials.