Discriminant partial least squares (PLS-DA)—a de facto standard classification method—was found to behave poorly when 3 classes of tequilas were modeled to study a collection of 170 commercial Mexican spirits measured by UV-Vis spectroscopy. This result was compared with other linear and nonlinear supervised classification methods (PLS with variable selection by SRI index and genetic algorithms; kernel-PLS—modified in this paper to handle simultaneously several classes, quadratic discriminant analysis (QDA), support vectors machines, and counter-propagation artificial neural networks). All linear models performed worse than nonlinear ones, and this was attributed to the quite different inner dispersion of the classes and the intermediate position of 1 class. Considering the overall classification results and parsimony, QDA was selected for routine assessments thanks to its simplicity and broad availability.
Andrade, J., Ballabio, D., Gomez-Carracedo, M., Perez-Caballero, G. (2017). Nonlinear classification of commercial Mexican tequilas. JOURNAL OF CHEMOMETRICS, 31(12) [10.1002/cem.2939].
Nonlinear classification of commercial Mexican tequilas
Ballabio, D;
2017
Abstract
Discriminant partial least squares (PLS-DA)—a de facto standard classification method—was found to behave poorly when 3 classes of tequilas were modeled to study a collection of 170 commercial Mexican spirits measured by UV-Vis spectroscopy. This result was compared with other linear and nonlinear supervised classification methods (PLS with variable selection by SRI index and genetic algorithms; kernel-PLS—modified in this paper to handle simultaneously several classes, quadratic discriminant analysis (QDA), support vectors machines, and counter-propagation artificial neural networks). All linear models performed worse than nonlinear ones, and this was attributed to the quite different inner dispersion of the classes and the intermediate position of 1 class. Considering the overall classification results and parsimony, QDA was selected for routine assessments thanks to its simplicity and broad availability.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.