Classification models for high-throughput electronic band structures using feature extraction

Applying machine learning techniques to aid materials engineering has become possible due to the existence of large databases of materials data. While models can be built using only scalar data, these databases also provide higher-dimensional data, such as high-throughput electronic band structures, which contain important information for many applications. For instance, the efficiency of thermoelectric materials is known to benefit from certain features in the band structure. We demonstrate how feature extraction techniques can be used to build classification models for band structures and show how these models can be used to aid materials innovation.