Data Mining: Applications to Nondestructive Assay Data

Data mining is the 1990s term for discovering useful information in large (usually high-dimensional) data sets. The availability of fast and inexpensive computers has resulted in widespread use of many computer-intensive data-mining procedures. Some data-mining practitioners have thrown caution to the wind; consequently, misleading performance claims abound. However, with appropriate corrections for the effect of choosing the best data model from among a large class of candidate models, there is much to be gained by mining data sets for information. This paper focuses on data-mining techniques that are well suited to enhance the performance and understanding of nondestructive assay methods for the assay of special nuclear material. There have been a few recent applications of data-mining techniques to NDA, so it is timely to review their limitations and benefits. We first give a brief history of the data-mining methods that are most suited for NDA applications. We next present analysis results for a subset of a large data set of neutron counting) of large containers with known amounts of plutonium in known locations in known matrices. We conclude with a brief review of other candidate NDA applications for data-mining tools.