Page 79 -

66 3 Data Clusterinrr

shown in Figure 3.13b. It is clearly visible that Factor1 is highly correlated with all
features except N and the opposite happens with Factor2. These observations
suggest, therefore, that the cork stoppers classification can be achieved either with
these two factors or with feature N and one of the other features.

FACTOR Principal
ANALYSIS components

121 ] -. 242
N 090 -.576
PRT .I13 -.383
ARM ,106 .296
PRM ,108 .246
ARTG ,123 .065
NG ,123 -.OD9
PRTG .I26 .037
RAAR .I10 .241
RAN 116 246

Figure 3.12. Dimensionality reduction of the first two classes of cork stoppers
using two eigenvectors. (a) Eigenvector coefficients; (b) Eigenvector scatter plot.

0.6
04. ARM
tR.M R:N
02. RAAR
AR:G PRTG
.. NG
rn 0.0
9
p -0.2
ART
0
-0 4
P!T
06
N
-0.8
066 0.72 0.78 0.84 0.90 0.96 '
b Factor 1
Figure 3.13. Factor loadings table (a) and graph (b) for the first two classes of cork
stoppers.

Notice that the scatter plot of Figure 3.12b is in fact similar to the one that
would be obtained if the data of Figure 3.11a were referred to the orthogonal
system of the factors.