gems, corundum, origin determination

Determining the geographic origin of ruby and sapphire is vital for their valuation, authentication, and ethical sourcing, yet traditional methods relying on expert assessment of visual, chemical, and spectroscopic traits can be subjective and inconclusive when deposits share overlapping features. This study applies machine learning to trace-element geochemical data from rubies and sapphires analyzed by EDXRF and LA-ICP-MS. Models trained on LA-ICP-MS data, with its superior sensitivity, achieved markedly higher classification accuracy than those based on EDXRF. Among the algorithms tested, random forests and artificial neural networks exceeded 90% accuracy for both ruby and sapphire, though metamorphic sapphires remained the most difficult to classify due to overlapping chemical profiles. The findings highlight machine learning’s potential to revolutionize gemmology by providing a scalable, reproducible, and objective framework for origin determination, paving the way for more standardized gem certification.