Segmentation Applications in Banking - Data Mining Techniques in CRM: Inside Customer Segmentation - page 234

Figure 6.2 The IBM SPSS Modeler stream for clustering.

purchase amount per merchant since, after a few trials, it became evident that
they produced a clearer separation which better reflected the relative spending
preferences, adjusting for the differences between customers in terms of their
overall spending amounts.
It is generally recommended that a data reduction technique, like principal
components or factor analysis, be applied to the input fields before clustering.
This approach was followed in the project presented here. The 14 fields of the
percentage of purchases by merchant type were used as inputs to a PCA model
which extracted a set of eight components, based on the ''Eigenvalues over 1''
criterion and a Varimax rotation. These eight components accounted for about
70% of the information carried by the original inputs, as shown in the ''variance
explained'' table, Table 6.3.

Table 6.3 Deciding the number of extracted components by examining the ''variance
explained'' table.

Total variance explained

Components Eigenvalue Percentage of variance Cumulative %

1 1.52 10.86 10.86

2 1.34 9.57 20.43

3 1.31 9.36 29.79

4 1.25 8.93 38.71

5 1.19 8.50 47.21

6 1.12 8.00 55.21

7 1.04 7.43 62.64

8 1.02 7.29 69.93

9 0.95 6.79 76.71