Data Mining Techniques for Segmentation - Data Mining Techniques in CRM: Inside Customer Segmentation - page 96

Table 3.11 ( continued )

Option Setting Functionality/reasoning for selection

• The initial eta refers to the model's learning rate
and the extent of change and adaptation of the
weights to the characteristics of the ''won'' record.
The model starts with coarse estimates and large
weight changes which are gradually decreased to
refine the solution. Thus, the learning rate at
phase 1 starts with the initial eta value speci-
fied (0.3) and (linearly or exponentially) decreases,
until completion of the specified phase 1 cycles, to
the initial eta value specified for phase 2 (0.1). Sim-
ilarly, the learning rate for phase 2 starts with the
value specified and decreases to 0 (convergence)
by the completion of phase 2 cycles

Users can experiment by setting the initial eta
values, for both phases, a bit higher, increasing the
sensitivity of the output neurons and permitting
larger adaptations to the presented data

EXAMINING AND EVALUATING THE CLUSTER SOLUTION

Analysts should examine and evaluate the revealed cluster solution and assess,
among other things, the number and relative size of the clusters, their cohesion,
and separation. A good clustering solution contains tightly cohesive and highly
separated clusters. More specifically, the solution, through the use of descriptive
statistics and specialized measures, should be examined in terms of the following.

THE NUMBER OF CLUSTERS AND THE SIZE OF EACH CLUSTER

Cluster examination normally begins by looking at the number of revealed clusters.
In general purpose clustering, analysts typically expect a rich but manageable
number of clusters. Analysts should look at the number of records assigned to each
cluster. A large, dominating cluster which concentrates most of the records may
indicate the need for further segmentation. Conversely, a small cluster with a few
records merits special attention. If considered as an outlier cluster it could be set
apart from the overall clustering solution and studied separately.