This study developed a robust method combining metabolomics and machine learning (ML) to authenticate the geographic origin of Wuyi rock tea, a premium oolong tea. Volatiles of 333 tea samples were profiled using gas chromatography time-of-flight mass spectrometry. Multilayer Perceptron achieved the best performance with an average accuracy of 92.7% on training data. The model showed over 90% accuracy on independent test sets. Gradient Boosting yielded the best accuracy (89.6%) using only 30 volatile features. This methodology shows promise for broader applications in identifying geographic origins of other agri-food products.
