One class self-organizing maps with Monte Carlo permutation for gemstone classification using Raman spectroscopy

Accurately identifying gemstones is important for authentication and quality control. In this study, we developed an artificial intelligence method that uses self-organizing maps (SOMs) combined with Monte Carlo testing to classify gemstones from their Raman vibrational patterns. The model was optimized for map size and training cycles, and its reliability was evaluated through repeated random tests. Raman data from the RRUFF database and real garnet samples were used for validation. The method performed extremely well, correctly recognizing gemstone types while avoiding false positives. Because it can also reject unknown samples instead of forcing them into incorrect classes, this SOM–Monte Carlo approach offers a reliable and robust tool for gemstone identification, even when reference data are limited.