Database Reference
In-Depth Information
example presented here, all communalities are quite high so there should be no
concern about the ''representativeness'' of the solution.
PROCEEDING TO THE NEXT STEPS WITH THE COMPONENT SCORES
The extracted components can be used in upcoming data mining models, provided
of course that they comprise a conceptually clear and meaningful representation of
the original fields. The PCA algorithm derives new composite fields, named com-
ponent scores, that denote the values of each record in the revealed components.
Component scores are produced through linear transformations of the original
fields, by using coefficients that correspond to the loading values. They can be used
as any other fields in subsequent tasks.
The derived component scores are continuous numeric fields with standard-
ized values; hence they have a mean of 0 and a standard deviation of 1 and they
designate the deviation from the average behavior. More specifically, the scores
denote how many standard deviations above or below the overall mean each record
lies. The list of the five component scores produced by PCA for 10 of the customers
in our example is shown in Table 3.7.
The high score of customer 5 in component 1 denotes a customer with
above-average SMS usage. The negative score in component 2 indicates low voice
usage. Similarly, customer 2 seems to be a person who frequently uses their phone
abroad (roaming usage measured by component 3) and customer 4 seems like a
typical example of a ''voice only'' customer.
As noted above, the derived component scores, apart from being fewer in
number than the original fields, are standardized, uncorrelated (due to Varimax
Table 3.7
A list of derived component scores.
Customer Component Component Component Component Component
ID
score 1
score 2
score 3
score 4
score 5
−
−
−
1
0.633
0.182
0.263
1.346
0.209
−
−
−
−
2
0.964
0.500
8.805
0.090
0.036
−
−
−
−
−
3
0.501
0.381
0.196
0.197
0.063
4
−
0.501
1.677
−
0.272
−
0.305
−
0.055
5
3.660
−
1.041
−
0.385
−
0.596
−
0.084
6
−
0.450
0.720
0.433
−
0.251
−
0.056
7
1.249
−
0.276
1.043
−
0.384
−
0.028
8
−
0.695
0.192
−
0.204
0.461
−
0.117
9
−
0.902
−
0.959
0.247
2.265
−
0.164
10
0.028
0.212
2.715
1.186
−
0.165
Search WWH ::
Custom Search