Data Mining in Materials Development

Abstract

Data Mining (DM) has become a powerful tool in a wide range of areas, from e-commerce, to finance, to bioinformatics, and increasingly, in materials science [1, 2]. Miners think about problems with a somewhat different focus than traditional scientists, and DM techniques offer the possibility of making quantitative predictions in many areas where traditional approaches have had limited success. Scientists generally try to make predictions through constitutive relations, derived mathematically from basic laws of physics, such as the diffusion equation or the ideal gas law. However, in many areas, including materials development, the problems are so complex that constitutive relations either cannot be derived, or are too approximate or intractable for practical quantitative use. The philosophy of a DM approach is to assume that useful constitutive relations exist, and to attempt to derive them primarily from data, rather than from basic laws of physics.