Characterization of Tobacco with Near-Infrared Spectroscopy with Competitive Adaptive Reweighted Sampling and Partial Least Squares Discrimination

Reconstituted tobacco products were characterized by near-infrared spectroscopy with classification models obtained through competitive adaptive reweighted sampling and partial least squares discrimination. Fifty samples were used to characterize the accuracy of the model. The results showed that all samples were classified correctly at the p > 5% level. The model provided excellent discrimination. The accuracy of training and prediction sets were 100 and 98.6%, respectively. The stability of the products were monitored by Hotelling T-2 statistics, and the results were satisfactory.