Segmentation Applications in Banking - Data Mining Techniques in CRM: Inside Customer Segmentation - page 238

Model Summary

Algorithm TwoStep

Input Features 8

Clusters 8

Cluster Quality

Poor Fair Good

− 1.0 − 0.5 0.0 0.5 1.0

Silhouette measure of cohesion and separation

Figure 6.3 The overall silhouette measure of the clustering solution for credit
card holders.

Table 6.6 Distribution of the derived clusters.

Percentage of credit card holders

Clusters Cluster 1 21.3

Cluster 2 10.0

Cluster 3 11.4

Cluster 4 9.1

Cluster 5 10.0

Cluster 6 25.4

Cluster 7 5.0

Cluster 8 7.7

Total 100.0

increased spending, while negative values denote scores below the average and
lower spending. In this particular example, though, we have to bear in mind
that negative scores in components 3, 4, and 5 represent increased spending for
appliances, telecommunication, and health services respectively, due to the high
negative loadings of the original inputs with those components.
Additionally, the clusters were profiled in terms of the original purchase fields.
The two affinity tables that follow, Tables 6.7 and 6.8, summarize the percentage
and the absolute purchase amount of each cluster by merchant type.