Using Quantile Comparisons to Classify Environmental Samples

Multivariate quantile comparisons (QC) is a method ideally suited for classification of particle samples because it is based on comparing different populations in which multiple samples are available. The basic procedure was presented at a regional INMM meeting and it has been developed to include consideration of measurement uncertainties, probability of misclassification, and to deliver a “none-of-the-above” classification decision. Environmental samples typically include measurements of elemental or isotopic inventories that include assessments of measurement error. The QC method incorporates the error by assuming it is normally distributed and repeatedly sampling from the error distribution. This approach fits naturally with the QC method, which uses repeated sampling from known classes and compares them to an unknown test sample, yielding a comparison score describing the similarity of the test sample to each class. Unlike many standard classification schemes, the QC method can recognize that none of the known classes adequately match the unknown test sample. This work considers an example problem that utilizes simulations of whole-core nuclide generation in a gas-cooled reactor. Samples taken at different times include different distributions of various nuclides and should be able to predict reactor burnup. This capability was demonstrated by previous research. Results incorporating measurement error into the simulated data indicate that very little classification capability is lost with small errors of 1-5%, but as measurement error increases to 10%, 20%, or 40%, the ability of the QC method to deliver accurate classification decisions is severely compromised.