Food products bearing a quality label such as organic production or geographical origin are generally more expensive than conventional products. They are therefore particularly exposed to the risk of fraud. Erroneous labeling of food origin can be detected by the measurements combining trace elements concentrations and the isotopic composition of the product (different isotopes depending on the use cases). However, these analyses tend to be quite expensive. We therefore seek approaches that require only a small number of isotopes to make reliable food origin prediction.

In computer science terms, we aim for accurate classification approaches that minimize the cost for the required feature extraction. More precisely, we are interested in understanding the trade-off between the cost of the chemicals and isotope computation and the gain in classification accuracy by including this data.

We are particularly interested in explainable models, to ease the discussion with relevant stakeholders such as food and nutrition authorities and food producers, and to gain insight into the temporal evolution of feature (i.e., isotope) importance. First approaches will be developed for specific case studies (e.g., wheat, wine, and cheese products). Once a classification and feature selection pipeline is established, we will apply this approach to other food products such as  spices, tea, or olive oil.

PhD student: Yihang LU

PhD supervisors: Dr Mathieu SEBILO (Director), Dr Carola DOERR (Co-director)

Research laboratory: IEES – UMR 7618 – SU-CNRS-IRD-INRAE-Paris Cité-UPEC - Equipe Ecologie integrative, des mécanismes aux services écosystèmiques