Identification of Cigarette Authenticity by CHAID Decision Tree Model Based on Distribution of Fluorescent Brightener in Different Cigarette Base Papers

The title method was proposed in view of the fact that the domestic and international standards prohibited the use of bis (triazine) aminostilbene fluorescent brightener in the cigarette base paper whereas the safety of the base paper of fake cigarette was not under control. The D65 fluorescent brightness and CIE whiteness (10 variables) of the 19 kinds of genuine and fake cigarette base paper samples (packaging paper, sealing paper, inner liner paper, filter plug wrapping paper and cigarette paper) were detected by whiteness meter. It was found that the CIE whiteness of various base papers of genuine and fake cigarette samples had no obvious pattern. The base paper with the D65 fluorescence brightness of genuine cigarettes not less than 1.00% was only cigarette paper (abnormal phenomenon, which needed further analysis). The base paper with the D65 fluorescence brightness of fake cigarettes not less than 1.00% included inner liner paper, sealing paper, cigarette paper and packaging paper, with the number ratios of 68.4%, 57.9%, 57.9% and 52.6%, respectively. SPSS 25.0 statistical software was used to establish a forward stepwise binary logistic regression model based on maximum likelihood estimation. The correlation analysis between the above 10 variables and the authenticity of cigarettes was carried out. After three-step regression, the 3 key variables were extracted, including the fluorescent brightness of the inner liner paper, the fluorescent brightness of the packaging paper, and the fluorescent brightness of the sealing paper. Based on the data of the 3 key variables of the samples over the years, classification model based on the Chi-square automatic interactive detection decision tree was established, and the accuracy of distinguishing the authenticity of cigarettes reached 100%. The infrared spectrum analysis of genuine and fake cigarette samples showed that the characteristic absorption peaks of these samples were basically the same, but the absorption peaks near 1 371 cm^-1 were observed in the infrared spectra of the fake cigarette samples, which might be derived from the bending vibration of the methyl C-H bond in the bis (triazine) aminostilbene fluorescent brightener. It was shown that the decision tree model established based on the distribution of fluorescent brightener could effectively identify the authenticity of cigarettes.