Data Mining Techniques for Segmentation - Data Mining Techniques in CRM: Inside Customer Segmentation - page 130

Figure 3.31 IBM SPSS Modeler recommended CHAID Stopping Criteria.
• Although specific clustering techniques offer automatic clustering methods for
selection of the clusters to fit, it is recommended to experiment and try different
solutions before approving the one for deployment.
• The evaluation of the clustering solution should involve an examination of the
separation and of the homogeneity of the derived clusters. Derived clusters
should present minimum intra-cluster and maximum inter-cluster variation.
But most of all, they should be interpretable and evoke business opportunities.
They should also correspond to recognizable customer ''types'' with clearly
differentiated characteristics. A clustering solution is justified only if it makes
business sense and if it can be used for the ''personalized'' handling of cus-
tomers, for instance for developing new products or for making new offers
tailored to the characteristics of each cluster. That is why data miners should
work together with the involved marketers in selecting the optimal clustering
solution.
• Profiling is an essential step in the evaluation of the clustering solution. By
using statistics and charts, even with the use of auxiliary models, data miners
should recognize the differentiating characteristics of each cluster. Each cluster