Machine learning is gaining attention as a new material discovery approach since it is impossible to explore the whole materials space, which is generated by combination of material structure and its constituent elements. Machine learning, which learns a lot of data on its own to find hidden rules, is expected as a methodology that can reduce the material space search speed and find materials beyond human intuition. To apply this to the development of new materials, it is essential to express the materials numerically and interpret the discovered rules as human-understandable knowledge. In this study, expression and interpretation methods for applying machine learning to materials science were studied. In the first chapter, we used machine learning to predict the degree of interaction between the material and oxygen atoms from the electronic structure of the material surface, and analyzed what aspects of the surface electronic structure were paid attention to when machine learning predicting by adding an attention mechanism. In the second chapter, we studied a machine learning method to reconstruct the three-dimensional structure of a material from X-ray diffraction patterns, and showed that the X-ray diffraction patterns can be used as a method representing the structure of a material.