Machine learning prediction of materials properties from chemical composition: Status and prospects

Abstract

In materials science, machine learning (ML) has become an essential and indispensable tool. ML has emerged as a powerful tool in materials science, particularly for predicting material properties based on chemical composition. This review provides a comprehensive overview of the current status and future prospects of using ML in this domain, with a special focus on physics-guided machine learning (PGML). By integrating physical principles into ML models, PGML ensures that predictions are not only accurate but also interpretable, addressing a critical need in the physical sciences. We discuss the foundational concepts of statistical learning and PGML, outline a general framework for materials informatics, and explore key aspects such as data analysis, feature reduction, and chemical composition representation. Additionally, we survey the latest advancements in the prediction of geometric structures, electronic properties, and other material characteristics from chemical formulas. The review also provides resource tables listing essential databases, tools, and predictors, offering a valuable reference for researchers. As the field rapidly expands, this review aims to guide future efforts in harnessing ML for materials discovery and development.