Honey Differentiation Using Infrared and Raman Spectroscopy Analysis and the Employment of Machine-Learning-Based...

Due to rising concerns regarding the adulteration and mislabeling of honey, new directives at the European level encourage researchers to develop reliable honey authentication models based on rapid and cost-effective analytical techniques, such as vibrational spectroscopies. The present study discusses the identification of the main vibrational bands of the FT-Raman and ATR-IR spectra of the most consumed honey varieties in Transylvania: acacia, honeydew, and rapeseed, exposing the ways the spectral fingerprint differs based on the honey's varietal-dependent composition. Additionally, a pilot study on honey authentication describes a new methodology of processing the combined vibrational data with the most efficient machine learning algorithms. By employing the proposed methodology, the developed model was capable of distinguishing honey produced in a narrow geographical region (Transylvania) with an accuracy of 85.2% and 93.8% on training and testing datasets when the Trilayered Neural Network algorithm was applied to the combined IR and Raman data. Moreover, acacia honey was differentiated against fifteen other sources with a 87% accuracy on training and testing datasets. The proposed methodology proved efficiency and can be further employed for label control and food safety enhancement.