In this podcast episode, things get analytical – we look at the idea that algorithms can predict the taste of wine. Numbers, data and models are replacing the sommelier – but how much truth is really in the system?
In the current episode, Jochen Fiedler and Esther Packullat from the Streuspanne team, together with guest Marcel Hensel, take a closer look at a study that promises to be able to predict wine quality using Machine Learning based on ingredients alone. Marcel is a postdoc at the Chair of Digital Farming at the University of Kaiserslautern-Landau (RPTU). Together, they show why the study's methodological approaches are a huge failure.
The idea sounds exciting: systematic, blind tastings meet comprehensive laboratory values. Numerous chemical parameters such as aromatic substances, colorants and pH value were measured in order to draw conclusions about the sensory quality of the wine. However, a closer look reveals tangible errors that unfortunately completely devalue the results.
What comes across here as scientific research is ultimately more of a nicely packaged statistical trap – and unfortunately is hardly suitable as a basis for reliable statements.
Listen in and uncork the truth with us!
Have you discovered a numerical phenomenon that you would like us to discuss? Or have you read a statistic in the media that you would like us to take a closer look at? Then get in touch with us at presse@itwm.fraunhofer.de!
The podcast can be found on all major portals such as Spotify, Podcasters, Amazon Music, Apple Podcasts, Deezer, RadioPublic, Pocket Casts or directly here on our website.
Fraunhofer Institute for Industrial Mathematics ITWM