From each sensor response, the classical feature (R/R0) was per-
formed and the PCA was applied to the data in order to reduce
its dimensionality, identify the relevant sensors in the array, and
discard redundant information. Subsequently, a graphic represen-
tation retain two principal components (PC1 and PC2) allow a much
easier interpretation of the data as shown in Fig. 4. In particular,
Fig. 4a shows the PCA plot obtained from the method (a), in pres-
ence of ethanol, and explains 98.18% of the data variability. This
PCA plot reveals a partial overlapping of the perfume samples. The
authentic perfume is not clearly separated from the inspired or
counterfeit perfume samples. Conversely Fig. 4b, corresponding to
the PCA plot obtained from the method (b) without ethanol, shows
a clear discrimination between the perfume samples with the
two principal components explaining 92.64% of the information.
All perfume samples are now well grouped and clearly separated
from each other. Two interesting features can be observed in this
plot. Firstly, there are a clear correlation between the price of the
inspired samples (see Table 1) and their proximity to the authentic
perfume. In this sense, inspired 3, 4 and 5 are progressively closer to
the original in clear contrast with the cheap inspired samples (1 and
2). On the other hand, the counterfeits are very far from the original
perfume, occupying a different region in the plot as compared to all
the inspired samples, which is indicative of a completely different