The chemometrics methods employed for classification are
supervised or unsupervised methods. The latter is usually referred
as clustering method too. The former needs the class property of
the training samples in addition to the classifier variables. Principal
component analysis (PCA) extracts the classifier features (named
principal components; PCs) from the spectroscopic data to visualize
the data and show the clustering pattern of different groups of
samples. Although, since the extracted PCs are obtained solely
from the spectroscopic data without using the class information,
the classification models constructed from the PCs usually represent
weak classification results. On the other hand, supervised
methods such as partial least square-discriminate analysis (PLSDA)
produce components that are more correlated with the data
classes. Though, PLS-DA suffers from poor performance in situations not unlikely to occur in real data. Canonical variates analysis
(CVA) is another supervised classification method,