Interpretable Machine Learning Applications: A Promising Prospect of AI for Materials

In recent years, data-driven machine learning has significantly advanced the design of new materials and transformed the research and development landscape. However, its heavy reliance on data and the “black-box” nature of its model-mapping mechanisms have hindered its application in materials science research. Integrating material knowledge with machine learning to enhance model generalization and prediction accuracy remains an important objective. Such integration can deepen the understanding of material mechanisms by screening physical and chemical features to uncover explicit intrinsic relationships. Thus, it promotes the advancement of materials science, representing a promising avenue for artificial intelligence (AI) applications in this field. In this review, the algorithms, functionalities, and applications in materials underlying interpretable machine learning approaches are summarized and analyzed. The impact of composition and microstructure on material properties is explored and mathematical expressions for intrinsic relationships of materials are developed. In addition, recent advancements in data- and knowledge-driven strategies for new material discovery, key property enhancement, multi-objective design trade-offs, and optimizing the entire preparation and processing workflow are reviewed. Finally, the future prospects and challenges associated with applying AI in materials science and its broader implications for the field are discussed.