Data Mining Techniques for Segmentation - Data Mining Techniques in CRM: Inside Customer Segmentation - page 129

Figure 3.30 IBM SPSS Modeler recommended CHAID Expert options.
Clustering techniques are applied for unsupervised segmentation. They
induce natural groupings of records/customers with similar characteristics. The
revealed clusters are directed by the data and not by subjective and predefined
business opinions. There is no right or wrong clustering solution. The value of each
solution depends on its ability to represent transparent, meaningful, and actionable
customer typologies.
Things to bear in mind about clustering:
• The clustering solution reflects the similarities and differences embedded in the
specific analyzed data. Therefore, the analysis should start with the selection of
the clustering fields, in other words the selection of the appropriate segmentation
criteria, which will best address the defined business objective.
• The clustering solution could be biased by large differences in the measurement
scales and by intercorrelations of the input fields. Therefore, a recommended
approach is to firstly run PCA to identify the distinct data dimensions and then
use the (standardized) component scores as inputs for clustering.