Data Mining and Multivariate Analysis in Materials Science

Abstract

Databases in materials science applications tend to be phenomenological in nature. In other words, they are built around a taxonomy of specific classes of properties and materials characteristics. In order for databases to serve as more than only a “search and retrieve” infrastructure, and more for a tool for “knowledge discovery”, data bases need to have functional capabilities. The recent advances in genomics and proteomics for instance provide a good example of the development of such “functional” databases. A first step to achieve this is to develop descriptors of materials properties that can be sorted and classified using appropriate data mining algorithms. In this paper we provide some examples of the use of some well established statistical tools to “prepare” such data especially when there is a multi-dimensional component associated with structure- chemistry-property relationships.