Construction and Verification of Classification Model for Flavor of Flue-cured Tobacco Based on Factor Analysis and Bayes Discriminant

In order to study the relationship between the chemical composition of flue-cured tobacco and its flavor, 500 tobacco samples from 71 cities/counties from of domestic provinces were collected from 2011 to 2013. Based on industrial standards and methods from the literatures standards we determined 114 chemical indicators which have been shown to have a significant impact on the quality of tobacco. The dimensions of each index were reduced using MFA (factor analysis), and the quantitative-Flavor factor scores were used in constructing and validating a Bayes discriminant model. The results showed that the original indicators can be made 22 common factors, which can explain 80.459 percent total variance of the original variables. Megastigmatrienone (A, C), His, anabasine, total cell wall material and other substances are widespread in tobacco leaves and can better represent their quality characteristics. Flavor of tobacco could be predicted based on the quantitative discriminant model, which is constructed by unsaturated aldehydes and ketones, acids, bases, and other substances in the cell wall material content, with the correct rate ≥ 83.3%. The model is easy to use and could be important in simplifying the process of tobacco flavor discrimination.