Untargeted metabolomic analysis of honey mixtures: Discrimination opportunities based on ATR-FTIR data and machine learning algorithms

Honey adulteration issues represent an important concern at distinct societal levels (i.e. producers, consumers, and state authorities) because honey represents one of the most falsified food commodities in the world. Adulterations can be more or less subtle and, as a consequence, these practices can be easy or very difficult to detect. One of the subtlest types of adulteration is represented by the detection of honey mixture, when honey is wrongly labelled as monovarietal. During the last few years, it was demonstrated that a refinement of the analytical results can be achieved by the employment of artificial intelligence in the development of food and beverages recognition models. In this light, our study proposes a new approach for the detection of colza honey addition to acacia one and the identification of the presence of sunflower honey in linden samples. For this purpose, the association between ATR-FTIR spectroscopy and machine learning algorithms was applied for recognition models development. Based on these models, it was possible to detect the mixture of colza-acacia mixture with an accuracy of 94.4%, while the blend of linden and sunflower honey was possible to be identi-fied with a 90.7% accuracy.

Automated summary

Honey adulteration is a significant concern globally, and the use of artificial intelligence in food recognition models has shown promise in improving detection accuracy. This study proposes a new approach using ATR-FTIR spectroscopy and machine learning algorithms to detect colza honey addition in acacia honey and identify the presence of sunflower honey in linden samples. The developed models achieved an accuracy of 94.4% in detecting colza-acacia mixture and 90.7% in identifying linden-sunflower honey blend.