Classification of vinegar types using volatile compound profiles and machine learning

Vinegar is a traditional fermented food whose aroma diversity arises from volatile metabolites formed during alcoholic and acetic acid fermentation. However, analyses of volatile profiles across vinegar types remain limited. Here, four vinegar types-rice, grape, apple, and persimmon-were characterized using headspace solid-phase microextraction coupled with gas chromatography-mass spectrometry (HS-SPME/GC-MS) and machine learning, identifying. 127 volatiles from 87 commercial samples. Among five models, Random Forest achieved the most stable performance (96.19% accuracy) Random Forest with Recursive Feature Elimination and Cross-Validation selected 21 predictive features while 18 corresponded to known fermentation-derived volatiles, 3 were identified as non-intentionally added substances or packaging-related contaminants. Shapley Additive exPlanations revealed type-specific drivers, including methyl acetate and 2-acetoxy-3-butanone (persimmon), acetic acid and isobutanol (grape), and benzaldehyde (rice). This framework provides an exploratory robust, data-driven approach for vinegar classification and the identification of dataset-specific volatile features, paving the way for future validation studies.