No abstract

Three models using partial least squares regression of near infrared reflectance spectroscopy were developed for the first time to evaluate the concentration of total alkaloids as nicotine, total nitrogen and total ash in powdered samples of Cuban cigar tobacco. They had lower standard error of cross-validation (SECV) and standard error of prediction (SEP) than those using: (1) principal components regression; (2) multiple linear regression and (3) other partial least squares models previously published for this and other tobacco types. SECV values in mass fraction units were: 0.0010, 0.0011 and 0.0048, respectively. The coefficients of determination (R2) in calibration were: 0.99, 0.94 and 0.96, respectively. Standard error of prediction values were: 0.0011, 0.0012 and 0.0049, respectively. The r2 obtained when comparing the predicted values using the validation set with the reference data was nearly the same as in calibration. The equation for total alkaloids as nicotine was good for quantification, and for process control, development and applied research according to the quotients range/standard error of prediction and standard deviation/standard error of prediction, respectively. Equations for the other analytes were as good as the one for alkaloids, but suitable only for screening purposes. Joint F-tests on the regression coefficients from ordinary least squares and iterative reweighted least squares fittings among predicted and reference data indicated slopes and intercepts enclosing 1 and 0, respectively. The relative systematic errors between the near infrared spectroscopy predicted values and the reference methods are close to the same figures reported previously for traditional analytical procedures under intermediate precision and reproducibility conditions; hence, the models are suitable for this industrial context.