Identification of uranium ore deposit type by rare earth elemental composition with multiclass classification using LDA (Linear Discriminant Analysis), QDA (Quadratic Discriminant Analysis) and KNN (K-Nearest Neighbours) pattern recognition techniques

Abstract Rare earth elements (REE) are considered to contain signatures for the provenance of uranium-bearing materials. Identifying the type of uranium deposit is an important part of nuclear forensic investigation. The identification involves matching the REE pattern of an intercepted material with a reference database. Since comparison databases are large, the matching requires multivariate data analysis techniques. In this paper, three pattern recognition techniques have been applied for the selection of a suitable approach for this purpose. These techniques include K-Nearest Neighbors, Linear Discriminant Analysis and Quadratic Discriminant Analysis. The models have been validated using bootstrap training and test splits. The reference database of the current study contains 497 samples covering 5 deposit types (Granite, Intrusive, Metamorphite, Sandstone-Roll Front and Unconformity-related) from 20 countries. Results have been evaluated using the sensitivity, specificity and likelihood ratios.