Key Points
- Researchers combined proton nuclear magnetic resonance (^1H-NMR) spectroscopy with machine learning models to authenticate wines by vintage, grape variety, and geographic origin with high accuracy.
- Logistic regression outperformed kNN, achieving over 98% accuracy in cross-validation and up to 100% in final tests, showcasing the method’s potential to combat wine fraud.
- The study offers a scalable, optimized workflow that could benefit wine producers, regulators, and consumers by ensuring product authenticity and enhancing confidence in wine labeling.
In a recent study published in the journal Beverages, a team of researchers from the National Institute for Research and Development of Isotopic and Molecular Technologies and Babeș-Bolyai University explored a new way to improve wine authentication (1). This study presents a new method that combines machine learning (ML) algorithms and proton nuclear magnetic resonance (1H-NMR) spectroscopy. By using this combined approach, the research team showed how this method could improve wine authentication because it can improve the classification of wine by vintage, cultivar, and geographical origin (1).
What is the state of the wine industry?
The wine industry exhibited sustained growth over the past 30 years, which has slowed because of recent modern-day economic developments (2). One of the main challenges the wine industry faces today is appealing to the younger generations as the Boomer population, which drives the majority of wine sales, declines (2). Despite these ongoing challenges, many wine owners report that the financial health of their businesses are good (2).
What was the experimental procedure?
As part of the experimental procedure, the team developed learning-based models to accurately differentiate wines based on key classification features using ^1H-NMR spectral data. With growing concerns over wine fraud and mislabeling, which can impact not only consumer trust but also the economic value of wines, advanced authentication methods are in increasing demand (1).
In their study, the research team implemented two machine learning (ML) models: k-nearest neighbors (kNNs) and logistic regression. These algorithms were trained on spectral data from a range of wine samples and tested for their ability to classify wines by vintage year, grape variety, and geographic origin (1). Given the complexity and high dimensionality of the spectral data, the researchers placed a particular emphasis on preprocessing. During the preprocessing steps, the researchers specifically selected the most relevant variables for each classification task (1).
Which ML-based model performed better?
The researchers found in their study that logistic regression performed better than kNN, achieving accuracy rates above 98% in 10-fold cross-validation procedures and up to 100% in final testing scenarios (1). This high degree of accuracy shows how the application of ML methods for food and beverage authentication is driving better results in this space.
Beyond performance metrics, the study offers practical recommendations for future applications. In particular, the authors proposed an optimized workflow for combining ^1H-NMR spectroscopy with machine learning techniques (1). This optimized workflow should include a preprocessing strategy that can include feature scaling and variable selection tailored to each classification type to allow for distinguishing wines by grape variety, vintage, or region of origin (1).
What are the implications of this study?
This study has implications for many individuals in the wine industry, including producers, regulatory agencies, and consumers. For producers, especially those in regions known for high-value wines, having a rapid and reliable authentication system could help deter counterfeiting and reinforce branding (1). For regulatory agencies, the combination of ^1H-NMR and ML could serve as a standard for verifying product labels, protecting both local economies and consumer safety (1). And finally, on the consumer front, this advancement promises greater confidence in the authenticity of the wines they purchase (1).
Furthermore, the integration of machine learning into traditional analytical workflows opens the door to automated, scalable solutions. As data sets grow and more wines are analyzed using ^1H-NMR, machine learning models can be continually updated and refined, offering real-time feedback and improved generalizability (1).
This study is part of a broader trend in food science and analytical chemistry that seeks to leverage artificial intelligence to solve long-standing challenges in quality control and traceability. It also exemplifies how interdisciplinary collaboration can result in new innovations to improve current beverage industry practices.
References
- Hategan, A. R.; Pirnau, A.; Magdas, D. A. Applications of Machine Learning for Wine Recognition Based on 1H-NMR Spectroscopy. Beverages 2025, 11 (2), 45. DOI: 10.3390/beverages11020045
- Silicon Valley Bank, State of the US Wine Industry Report 2025. Silicon Valley Bank. Available at: https://www.svb.com/trends-insights/reports/wine-report/ (accessed 2025-06-10).
