French Scientists Use Machine Learning to Authenticate Champagne and Burgundy Origins With 100% Accuracy

French researchers have developed a new method to authenticate the origin of sparkling wines using machine learning and geochemical fingerprinting. The study, conducted by scientists from Sorbonne Université and published in npj Science of Food, focused on 75 sparkling wine samples from the Champagne and Burgundy regions. The team combined isotopic and elemental analysis with advanced machine learning algorithms to distinguish between wines from these two renowned French regions.

The research addresses a growing problem in the global wine market: counterfeiting. High-value wines, especially those labeled as Champagne or Burgundy, are frequent targets for fraud. Traditional methods for verifying wine origin rely on supply chain documentation and geographical indications, but these can be manipulated or forged. Analytical techniques such as nuclear magnetic resonance spectroscopy and mass spectrometry have been used in the past, but their high cost and complexity have limited widespread adoption.

In this study, the researchers measured the strontium isotope ratio (87Sr/86Sr) in each wine sample. This ratio is influenced by the geology of the vineyard’s soil and is difficult to fake or alter during winemaking. Using logistic regression, a transparent machine learning model, they achieved 100% accuracy in classifying the wines’ origins based on this isotopic marker.

To address the high cost of isotopic analysis—about 300 euros per sample—the team also tested whether elemental concentrations could serve as reliable alternatives. They found that rubidium (Rb) concentration alone provided over 90% classification accuracy while reducing analytical costs by 75%. This makes routine authentication more feasible for producers and regulatory agencies.

The researchers used a dataset of 66 Champagne samples and 9 Burgundy samples. To ensure robust results despite this imbalance, they applied synthetic minority oversampling (SMOTE) and repeated cross-validation. They compared three machine learning models: logistic regression, random forest, and support vector machines. Logistic regression performed best, with an average F1-score of about 0.93.

Single-feature analysis showed that the strontium isotope ratio was the most powerful discriminator between regions, followed closely by rubidium concentration. Combining both features further improved classification performance. The study also explored correlations among different elements and isotopes, finding that some combinations captured unique aspects of wine composition linked to regional geology.

The practical implications are significant for the wine industry. By using rubidium as a proxy for more expensive isotopic measurements, wineries and regulators can implement large-scale authentication programs at lower cost. The method is also transparent and interpretable, which is important for regulatory acceptance.

The authors note that their approach could be extended to other high-value foods prone to fraud, such as olive oil, honey, or coffee. They acknowledge some limitations: the current dataset covers only two French regions, and factors like climate variation across vintages may affect geochemical signatures. Standardized protocols and reference materials will be needed for broader adoption.

Sample collection was carried out under strict cleanroom conditions in France. Elemental analysis was performed using quadrupole inductively coupled plasma mass spectrometry (Q-ICP-MS), while isotopic ratios were measured with multicollector ICP-MS (MC-ICP-MS). The workflow included sample digestion, matrix purification with ion-exchange resin to remove interferences, and calibration with certified standards.

The study was supported by the European Union’s Horizon Europe program and involved collaboration with Moët et Hennessy for access to authentic wine samples. The findings offer a scalable solution for protecting appellation integrity in an industry where consumer trust is closely tied to origin claims.

As online sales expand and counterfeit risks grow worldwide, this research provides a scientific basis for more secure wine authentication. By integrating analytical chemistry with machine learning, French scientists have taken a step toward ensuring that what is labeled as Champagne or Burgundy truly comes from those storied regions.