In authenticity checking of tobacco products, the identification of the varieties present is of primary importance. Nowadays the detection of illegal tobacco products is often based on package analysis and administrative verification, sometimes complemented with laboratory analysis.

In this study an approach based on IR spectroscopy (MID-IR and NIR), for the identification of tobacco varieties in tobacco blends is proposed. Therefore, different blends were prepared, spectra were measured and binary PLS-DA models were created. All models were evaluated and compared for their predictive performance, using both cross validation (internal validation) and an external test set. For the best performing model for each analyte the limit of detection was estimated. Finally, quantitative models were created to estimate the relative amount of one of the targeted varieties in the mixtures and a proof of concept using five commercial tobacco blends was performed.

NIR proved to outperform MID-IR with maximum values of correct classification rate, precision, specificity and accuracy for four varieties and only one misclassification for the two remaining ones. Indicative limit of detection values were obtained between 1 and 8%. Quantitative errors were all smaller than 5%. These values as well as the application to commercial samples proved the feasibility of the presented approach and its potential value as tool in the fight against fraud and counterfeited tobacco products.