N-SLOPE: A One-Class Classification Ensemble for Nuclear Forensics
Justin Kehl, Lubomir Stanchev · 2018
One-class classification is a specialized form of classification from the field of machine learning. A traditional classifier always assigns a new element to one of the known classes, but it cannot handle elements that do not belong to any of the existing classes. One-class classification seeks to identify these outliers, while still correctly assigning the rest of the elements to classes appropriately. One-class classification is applied here to the field of nuclear forensics, which is the study and analysis of nuclear material for the purpose of nuclear incident investigations. Nuclear forensics data poses an interesting challenge because false positive identification can prove costly and the data is often small, high-dimensional, and sparse, which is problematic for most machine learning approaches. A web application that incorporates N-SLOPE (a machine learning ensemble) is built using the R programming language and the shiny framework. N-SLOPE combines five existing one-class classifiers with a novel one-class classifier called SCD (Soft Centroid Distance) algorithm and uses ensemble learning techniques to combine output. NSLOPE is validated on an enhanced version of Galaxy Serpent 3, which is a recent international nuclear forensics exercise. NSLOPE achieves high classification accuracy of 85% on this difficult data set, while minimizing false positive detection rate to zero by correctly detecting all 16 outliers that are present in the data set.