SAIP2026

XRF-derived major elements for nuclear forensic signatures in radiothermic carbonatites area using machine learning and multivariate modelling

7 Jul 2026, 11:20

20m

Lecture Hall GH2 (University of the Western Cape)

Oral Presentation Track B - Nuclear, Particle and Radiation Physics Nuclear, Particle and Radiation Physics -1

Speaker

Liteboho Ntsohi (University of the Witwatersrand)

Description

This study evaluates the potential of major elemental compositions obtained from X-ray fluorescence (XRF) spectrometry (11 XRF majors: $SiO_{2}$, $K_2 O$, $Na_2 O$, $TiO_2$, $Fe_2 O_3$, $Al_2 O_3$, $P_2 O_5$, $CaO$, $MgO$, $NiO$, $MnO$) for characterizing uranium ore samples from the Mrima Hill carbonatite complex in Kenya, with the aim of assessing their suitability as nuclear forensic signatures. Given the compositional nature of major element data, centered log-ratio (CLR) transformation will be applied to ensure statistical validity. A machine learning framework comprising dimensionality reduction, unsupervised clustering, and classification algorithms, will be utilized to evaluate intra-deposit variability and assess the structure and consistency of geochemical signatures for forensic applications. The study further aims to quantify the extent to which variability within the deposit influences the stability and robustness of derived signatures. The analysis will focus on defining the multivariate structure of the dataset through mean vectors and covariance relationships, enabling future probabilistic attribution using distance-based metrics. The resulting characterization is intended to serve as a baseline dataset for integration into nuclear forensic databases. This work contributes to ongoing efforts in nuclear forensics by critically evaluating the capabilities and limitations of XRF-derived major elements as practical indicators of nuclear forensic provenance.