Development of a data-mining methodology for spent nuclear fuel forensics

The purpose of this study is to categorize the type of spent nuclear fuels using simulation data-based classification methods. Considering the practical conditions making the full analysis of radioactive nuclides difficult, the classification methods were designed to be robust to noise and missing information. The strength and weakness of three classifiers, linear discriminant analysis, quadratic discriminant analysis and support vector classification were compared, which is developed by the history information such as burnup, enrichment, and cooling type generated from ORIGEN-ARP upon fuel assembly types. Auto-Associative Kernel Regression improved outlier management as a pre-processing technique.