To improve the accuracy of coffee origin authentication, we developed a pyrazine-targeted fingerprinting approach that synergistically integrates high-performance liquid chromatography (HPLC), chemometrics, and machine learning for origin discrimination. After validating the HPLC fingerprinting method, we profiled 180 batches, identifying nine shared pyrazine peaks. Based on the proposed fingerprint, chemometric analyses partitioned the 180 batches into five clusters and identified five discriminant pyrazine components. Five machine-learning models were developed; the deep neural network (DNN) performed best, achieving 86.1% accuracy on an internal hold-out validation set, 88.33% mean accuracy in 5-fold cross-validation, and 85.0% accuracy on an independent external test set from a different harvest batch ( n = 20). Overall, an interpretable pyrazine fingerprint coupled with modern analytics enables efficient origin discrimination and supports authentication and supply-chain oversight. • HPLC fingerprints of nine alkyl pyrazines were constructed for origin verification. • OPLS-DA identified five pyrazine markers that strongly discriminate coffee origins. • DNN model achieved 100% accuracy, outperforming RF,RT,KNN and SVM. • Combining HPLC and deep learning enables rapid coffee authentication.
Lian et al. (Sun,) studied this question.