Tensor-Based K-Nearest Neighbors Approach for Enhanced Neutron–Gamma Discrimination

Neutron detection is essential in nuclear safety, medical diagnostics, and high-energy physics. However, reliably discriminating neutrons from gamma rays remains a major challenge in mixed radiation fields. This work investigates the performance of a K-Nearest Neighbors (KNN) classifier applied to features extracted through two Tucker decomposition–based signal separation methods: Multiway Blind Source Separation (MBSS) and Sequentially Truncated Multilinear Singular Value Decomposition (ST-MLSVD). The approach yielded high discrimination performance, with all tested configurations achieving a Figure of Merit (FOM) ranging from 1.61 to 3.91 and an F1 score of 100%. In particular, the MBSS–KNN and ST-MLSVD–KNN combinations in mode 2 produced the most reliable outcomes, combining robustness with strong accuracy for neutron–gamma classification. These results confirm the relevance of tensor-based feature extraction coupled with supervised learning, offering a promising pathway for reliable particle identification in complex radiation environments.