Machine Learning-Assisted Headspace Gas Chromatography Ion Mobility Spectrometry for Characterization and...

Perillae Folium, derived from the leaves of Perilla frutescens (L.) Britt., is recognized as both a medicinal substance and an edible ingredient in China, with its market demand steadily rising. The nutritional value of Perillae Folium varies markedly across different geographical origins, posing a constraint on its further industrial development. To date, there remains a paucity of robust and reliable approaches for tracing the geographical origin of Perillae Folium samples. This study employed headspace gas chromatography ion mobility spectrometry to profile VOCs in Perillae Folium, revealing that those originating from Hunan province contained comparatively higher volatile organic component levels. Subsequently, a total of 54 volatile organic components were putatively identified in Perillae Folium, and six featured VOCs including 2-decanone, 3-octanol, 2,6-dimethyl-5-heptenal, 3-methyl-3-buten-1-ol, ethyl 3-hydroxybutyrate, and 1-(1-methylethyl)-4-methylbenzene were identified. Eight machine learning classifiers were built based on the selected featured VOCs. Among the eight models, the linear and polynomial kernel support vector machines achieved the highest predictive accuracies, both exceeding 90%, indicating their strong potential for practical application. This study provides a reference for the study of volatile organic components in Perillae Folium from different geographical origins, and offers a simple, accurate, and reliable method for distinguishing Perillae Folium from these regions.