The whole data set (180 samples) was split into two sets: a
calibration set consisting of 126 samples (7 samples from each group)
and a prediction set consisting of 54 samples (3 samples from each
group). The PLSR models were built with the calibration set using a
full cross-validation method (leave-one-out), removing one observation
at one time from the sample sets until all samples have been
removed once. This is the method generally used in developing the
PLSR model for detecting a specific material with a spectral data set.