Akademik Veri Yönetim Sistemi
Journal of Food Process Engineering, cilt.49, sa.2, 2026 (SCI-Expanded, Scopus)
Honey crystallization studies are increasingly popular. Paradoxically, it is difficult to accept because it makes it harder to satisfy the consumer's senses. In this research, spectrum analysis was used to provide information about honey crystallization and its source using machine learning. All honey samples were maintained at +4°C during crystallization. The honey sample from Ankara city crystallized in almost two weeks, from Manisa city crystallized in 1 month, and from Artvin city crystallized in 2 months. The research measurement system consists of a shooting tent, a Celestron MicroDirect digital microscope, and Theremino spectral imaging software. Each sample was placed in a 9 × 1.5 cm Petri dish, with a 0.5 cm space from the digital microscope. Each honey sample in a Petri dish was then measured from 100 distinct sites, and the average measurement was recorded as the final data point for both pre- and post-crystallization states. All honey samples yielded a total of 1.634.400 spectral measurements. As a result, it has been determined that honey samples from Artvin city (Black Sea), Ankara city (Central Anatolia), and Manisa city (Aegean Region) of Türkiye can be separated based on wavelength and light intensity percentiles by using k-Means, k-Medoids, and Fuzzy c-Means clustering algorithms in the KNIME analytics platform. The best results were found as: 0.999 (k-Means), 0.995 (k-Medoids), 1 (Fuzzy c-Means) as control, and 0.998 (k-Means), 0.986 (k-Medoids), 0.998 (Fuzzy c-Means) during crystallization, and 0.999 (k-Means), 0.991 (k-Medoids), 0.998 (Fuzzy c-Means) as crystallized, crystallization clustering quality values, respectively.