Premium
Multivariate classification of Southern Brazilian table wines
Author(s) -
Hansen Lucas,
Ferrão Marco Flôres
Publication year - 2020
Publication title -
journal of chemometrics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.47
H-Index - 92
eISSN - 1099-128X
pISSN - 0886-9383
DOI - 10.1002/cem.3302
Subject(s) - principal component analysis , artificial intelligence , pattern recognition (psychology) , linear discriminant analysis , multivariate statistics , computer science , test set , partial least squares regression , data set , statistics , mathematics , machine learning
Abstract In this work, we evaluated the performance of several classifiers (supervised Kohonen self‐organizing maps (KSOMs), soft independent modelling of class analogy (SIMCA), k‐nearest neighbors (kNN), and partial least squares with discriminant analysis (PLS‐DA) in the multiclass classification of Southern Brazilian table wines based on their physicochemical data. We also employed an unsupervised KSOM for the exploratory analysis of our data and compared its performance to that of PCA in this same task. All methods tested here presented a non‐error rate (NER) and accuracy equal to or higher than 67% in the classification of the samples, having PLS‐DA achieved an NER of 86% in classifying the samples from the test set and accuracy of 83% in classifying the samples from the training set. However, the best overall classification performance (when classification performances in training, cross‐validation, and test sets are taken into account) in terms of NER and accuracy was that of SIMCA. Regarding the unsupervised analysis of the data, principal component analysis (PCA) provided a better separation of the samples and more convenient visualization of relationships between variables, and between variables and samples, than unsupervised KSOMs.