Data Mining Techniques for Segmentation - Data Mining Techniques in CRM: Inside Customer Segmentation - page 82

Table 3.8 IBM SPSS Modeler recommended PCA options.

Option Setting Functionality/reasoning for selection

Extraction
method

PCA PCA is the recommended data reduction algorithm in

cases where data reduction is the first priority
Worth trying alternatives: other data reduction algo-
rithms like principal axis factoring and maximum
likelihood are the most commonly used factor analysis
methods

Extract
factors

Eigenvalues
over 1

Themost common criterion for determining the number
of components to be extracted. Only components
with eigenvalue above 1 are retained
Worth trying alternatives: users can intervene by
setting a different threshold value. Alternatively,
by examining alternative extraction criteria, they
can specify a specific number of components to
be retained, through the ''Maximum number (of
factors/components)'' option

Rotation Varimax Varimax rotation can facilitate the interpretation of the

components by producing clearer loading patterns.
It is an orthogonal rotation method which produces
uncorrelated components

Sort values Selected Aids the interpretation of components by improving

the readability of the rotated component matrix.
It gathers together the original fields which are
associated with the same component

Hide values
below

0.3 Aids component interpretation by improving the

readability of the rotated component matrix. It
suppresses loadings with absolute values less than 0.3
and allows users to focus on significant loadings

CLUSTERING TECHNIQUES

In this section we will examine some of the most well-known clustering techniques.
Specifically we will present the K-means, the TwoStep, and the Kohonen network
algorithms. Moreover, since a vital part of a segmentation project is insight into the
derived clusters and an understanding of their meaning, we will also propose ways
for profiling the clusters and for outlining their differentiating characteristics.