Quantitative descriptive analysis and electronic nose combined with machine learning methods to quickly predict consumer preference for Cheddar cheese

Consumer preference for Cheddar cheese is influenced by its aroma characteristics, with aroma serving as a key sensory attribute determining consumer acceptance. This study innovatively integrates quantitative descriptive analysis (QDA) with electronic nose (e-nose) technology to systematically characterize the flavor profiles of 27 common Cheddar cheeses available in the Chinese market. We employed QDA to evaluate six odor attributes of cheese, utilizing e-nose technology to rapidly capture volatile flavor compounds. Using the obtained QDA and e-nose datasets, 4 machine learning algorithms, namely, logistic regression, Gaussian support vector machine, k-nearest neighbors, and decision tree, were used to classify and predict the preference of Chinese consumers for Cheddar cheese. The study recruited 152 consumers, who categorized their preferences into "like," "neutral," and "dislike" based on preference intensity. Principal component analysis revealed that the first 2 principal components of the QDA and e-nose data cumulatively explained 58.5% and 84.3% of the total variance, respectively. When using the full feature set, the classification accuracies of the 4 machine learning models were 87.5%, 87.5%, 75.0%, and 81.2%, respectively. After feature optimization (excluding QDA indicators), model performance significantly improved, achieving accuracies of 87.5%, 92.9%, 87.5%, and 92.9%, respectively. This study confirms the effectiveness and application potential of combining sensory omics with machine learning methods in predicting cheese consumption preferences. It provides key technical support for the precise development of cheese products tailored to Chinese consumers' flavor preferences and holds significant practical implications for implementing flavor-directed regulation in the dairy industry.