Data Mining Approaches to High-Throughput Crystal Structure and Compound Prediction

Predicting unknown inorganic compounds and their crystal structure is a critical step of high-throughput computational materials design and discovery. One way to achieve efficient compound prediction is to use data mining or machine learning methods. In this chapter we present a few algorithms for data mining compound prediction and their applications to different materials discovery problems. In particular, the patterns or correlations governing phase stability for experimental or computational inorganic compound databases are statistically learned and used to build probabilistic or regression models to identify novel compounds and their crystal structures. The stability of those compound candidates is then assessed using ab initio techniques. Finally, we report a few cases where data mining driven computational predictions were experimentally confirmed through inorganic synthesis.