Similarity evaluation of cigarette aroma components based on kernel principal component analysis

The type and content of aroma components are the main factors affecting the tobacco style. For tobacco production enterprises, the style analysis is very important to compare their own products with the competitors. Principal component analysis method based on kernel function is a nonlinear multivariate statistical analysis method with favorable effect in dimensionality reduction. In this paper, the high-dimensional data up to 63 index of tobacco aroma compositions were mapped into two-dimensional feature space by kernel principal component analysis, and the information amount of feature data is not less than 95%. Inter-class dispersion was introduced to evaluate the aroma components similarity of nine brands products. The results showed that the evaluation of nine brands cigarette based on the proposed method was consistent with the Chinese cigarette style feature classification and brand market segmentation. Such method can provide effective theoretical support for the traditional classification method of cigarette-style feature.