Abstract: Multivariate statistical methods were used to establish a flavor model of fermented milk. The parameters of major flavor compounds such as carbonyl compounds (aldehyde and diacetyl), acids (volatile acid and total acidity), and fermented milk components (sugar, fat, protein and non-fat milk solids) were analyzed as independent variables, and sensory coordination of flavor as the dependent variable. In order to select the indices with great impact on dependent variable and little correlation with independent variables, multiple linear regression and principal components analysis (PCA) were conducted to analyze 8 flavor indices. Finally, the concentrations of aldehyde, volatile acid, sugar and fat were selected as the independent variables and sensory evaluation score of flavor coordination as dependent variable. The coefficients of variables in the established model revealed statistical significance and weak multicollinearity was observed. The adjusted R2 of the final model was 0.447. The determined values and the predicted values of the tested samples were fitted well with an average error of 7.18%.

Key words: fermented milk, flavor, multiple linear regression, principal components analysis (PCA), modeling