Multivariate chemometric discrimination of cigarette tobacco blends based on the UV-Vis spectrum of their hydrophilic extracts

The application of UV-Vis spectrophotometry as an alternative or complementary approach to the classification of tobacco products is presented in this work for the first time. Two hundred fifty samples from five different cigarette brands composed of single and mixed tobacco blends were examined for that purpose on the basis of the UV-Vis spectrum of their aqueous extracts. Data transformation based on the normalization of absorbance intensities as a function of sample weight was employed in order to account for differences in the relative intensities of each sample. Principal components analysis (PCA) was used to extract outlier cases and sample classification was then pursued with the aid of discriminant analysis (DA) suggesting that a reduced number of variables (thirteen out of seven hundred initially available) could provide perfect classification (100% correct assignations) of samples containing single tobacco species or different blends and a fair classification of samples with similar composition (80% correct assignations) yielding an overall 95.7% correct classification. To this pursue, classification and regression trees were found to afford perfect classification of all samples using only a few logic rules based on appropriate split conditions at the expense of inserting 15 variables in the model.