Discrimination of vineyard origin and grape cultivars using rapid mass spectrometry and machine learning

• Rapid LC-qTOF MS combined with machine learning enables grape authentication. • PCA identified as the most effective dimensionality reduction approach. • Thirty principal components outperformed the 16,521 original spectral features. • The PCA–RF model achieved 95% accuracy for vineyard cultivar classification. Identifying grape cultivars and vineyard origins is essential for ensuring wine quality, traceability, and authenticity. This study presents a rapid liquid chromatography–quadrupole time-of-flight mass spectrometry (LC-qTOF MS) approach that substantially reduces analysis time compared with conventional methods. Mass spectral data of grape cultivars collected from 11 vineyards across New South Wales, Australia, within an m/z range of 45–1500 were analysed to discriminate cultivars based on their geographical origin. To address the high dimensionality of the dataset, feature extraction was performed using principal component analysis (PCA), supervised stacked autoencoder (SAE), and uniform manifold approximation and projection (UMAP). Among these dimensionality reduction techniques, PCA exhibited the most robust and consistent performance. Fifteen machine learning models were then evaluated to assess classification accuracy. The Random Forest (RF) model, when combined with PCA, achieved the highest accuracy (95.6%), effectively distinguishing grape cultivars from the 11 vineyard sites. Overall, these findings demonstrate that integrating rapid LC-qTOF MS with machine learning provides a powerful and efficient framework for authenticating grape cultivars and classifying vineyard origins, highlighting the potential of data-driven approaches for food provenance and quality assurance.