This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
,
Department of Physics, Faculty of Sciences and Mathematics, University of Nis , Niš , Serbia
Department of Physics, Faculty of Sciences and Mathematics, University of Nis , Niš , Serbia
Radiation measurements obtained with simple detectors are well described by Poisson counting statistics, where the mean count rate carries most of the relevant physical information. Here, we examine whether standard machine learning (ML) methods can improve the classification of radiation measurement scenarios beyond what can be achieved using basic statistical descriptors. Measurements were performed with a Geiger counter for seven distinct configurations, involving two radioactive sources of different strengths, varying source–detector distances, and shielding conditions. For each configuration, multiple samples were collected as short sequences of repeated count measurements. Several approaches were compared, including a nearest-mean classifier, Random Forests, and other standard classifiers (including distance-based and linear models). The results show that the simple nearest-mean approach consistently matches or outperforms the tested ML methods. Using the full sequence of counts or additional derived features did not lead to improved performance. This is consistent with the underlying Poisson statistics, for which the sample mean acts as a sufficient statistic for the count rate. These results illustrate a practical limitation of ML in settings where the data-generating process is well understood and dominated by simple statistical structure.
Ministry of Science, Technological Development, and Innovation of the Republic of Serbia through Agreement No. 451-03-34/2026-03/200124
The statements, opinions and data contained in the journal are solely those of the individual authors and contributors and not of the publisher and the editor(s). We stay neutral with regard to jurisdictional claims in published maps and institutional affiliations.
