Segmentation Applications in Banking - Data Mining Techniques in CRM: Inside Customer Segmentation - page 235

Table 6.3 ( continued )

Total variance explained

Components Eigenvalue Percentage of variance Cumulative %

10 0.84 6.00 82.71

11 0.73 5.21 87.93

12 0.70 5.00 92.93

13 0.59 4.21 97.14

14 0.40 2.86 100.00

As the 14 original fields represented a relatively high categorization level
of merchants, the benefits of PCA in terms of data reduction may not be
evident in this particular case. However, having to analyze tens or even hundreds
of fields is not unusual in data mining. In these situations a data reduction
technique can provide valuable help in understanding and ''tidying'' up the available
information.
But what do these eight derived components represent? Since they were
about to be used as the segmentation dimensions, it was crucial for the project
team to understand them. The components were interpreted in terms of their
association with the original fields. These associations (correlations or loadings) are
listed ''in the rotated component matrix'' given in Table 6.4.

Table 6.4 Understanding and labeling the components through the rotated
component matrix.

Rotated component matrix

Components

1 2 3 4 5 6 7 8

PRC APPAREL 0.73

PRC ACCESSORIES 0.66

PRC FOOD 0.74

PRC GAS 0.53

PRC APPLIANCES − 0.86

PRC TELCOS − 0.88

PRC HEALTH − 0.93

PRCHOMEGARDEN 0.91

PRC TRAVEL 0.44 0.32 0.82

PRC FITNESS 0.31

( continued overleaf )