Data Mining Techniques for Segmentation - Data Mining Techniques in CRM: Inside Customer Segmentation - page 78

Table 3.6 The communalities matrix.

Communalities

VOICE OUT CALLS 0.913

VOICE OUT MINS 0.901

SMS OUT CALLS 0.791

OUT CALLS ROAMING 0.838

GPRS TRAFFIC 0.762

MMS OUT CALLS 0.808

PRC VOICE OUT CALLS 0.953

PRC SMS OUT CALLS 0.954

PRC OUT CALLS ROAMING 0.847

PRC INTERNET CALLS 0.778

PRC MMS OUT CALLS 0.823

DOES THE SOLUTION ACCOUNT FOR ALL THE ORIGINAL FIELDS?

The next question to be answered is whether the derived solution sufficiently
accounts for all the original fields. This question is answered by Table 3.6, which
contains the communalities of the original inputs.
Communalities represent the amount of variance of each field that is jointly
accounted for by all the extracted components. High communality values indicate
that the original field is sufficiently explained by the reduced PCA solution. Analysts
should examine the communality table and assess whether the level of explained
information is satisfactory (typically above 0.50) for each input.

Technical Tips on Communalities

Communality represents the total amount of variance of a specific field that
is jointly accounted for by all the components. It is calculated as the sum of
squared loadings of the field across all components.

A low communality value designates an original field with low loadings on the
components and with an insignificant contribution to the formation of the PCA
solution. Thus communality values can help in the identification of original fields
left unexplained by the derived solution. Extraction of additional components may
help in situations like these as it may account for the ''forgotten'' inputs. In the