Data Mining Techniques for Segmentation - Data Mining Techniques in CRM: Inside Customer Segmentation - page 91

Table 3.10 ( continued )

Option Setting Functionality/reasoning for selection

The automatic clustering option is the typical approach
for starting model training
Worth trying alternatives: apart from the automatic
solution proposed by the algorithm, users can also
evaluate alternative solutions by imposing a different
range for automatic clustering. They can also ''take
the situation into their own hands,'' deactivate the
automatic clustering, and explicitly ask for a specific
number of clusters, by selecting the ''Specify number
of clusters'' option and setting the required number
of clusters to the respective text box

Exclude out-
liers

Selected This option enables handling of outliers, an option

generally recommended as it tends to yield richer
clustering solutions

Figure 3.8 IBM SPSS Modeler recommended TwoStep Model options.

CLUSTERING WITH KOHONEN NETWORK/SELF-ORGANIZING MAP

Kohonen networks are special types of neural networks used for clustering. The net-
work typically consists of two layers. The input layer includes all the clustering fields,