Rapid discrimination of geographical origin and analysis of chemical characterization of tobacco leaves from multiple countries

Tobacco is a globally cultivated crop featuring distinct quality variations among leaves from different geographical origins. To develop a rapid, robust, and accurate method for multi-origin traceability, this study employed near-infrared spectroscopy combined with rapid chemical composition analysis to obtain 70 chemical components in samples from nine major tobacco-producing regions in China and four other countries (the United States, Brazil, Zimbabwe, and Zambia). One-way analysis of variance (ANOVA) and hierarchical cluster analysis (HCA) were used to investigate regional chemical differences. Discrimination models were built using a support vector machine (SVM), a backpropagation neural network, and a random forest. The best model was interpreted using permutation feature importance (PFI) to identify key markers for origin discrimination. One-way ANOVA revealed significant differences (p ≤ 0.001), and HCA demonstrated clear regional patterns. The SVM-hybrid kernel achieved the best performance with 97.96% test accuracy and macro-average recall, precision, and F1 scores of 0.9836, 0.9806, and 0.9821, respectively. The PFI algorithm was employed to identify and rank the top 20 key chemical components influencing the geographical origin discrimination. The top ten key components were Fru-Asn, succinic acid, rutin, Fru-Val, sulfate, serine, phosphate, starch, potassium, and Fru-Gly. This study integrated chemometrics, near-infrared, rapid chemical analysis, and interpretable machine learning to accurately distinguish tobacco origins, reveal regional traits, and offer insights into geographical traceability and chemical profiling.