The first part of our methodology involved ranking the attributes
we had chosen for analysis. This ranking was done in order to
determine which attributes were relevant to the information we were
mining. Ranking enabled us to eliminate any attribute(s) that were of
little-or-no significance to our decision attribute.
Next, we looked at all client accounts to predict which accounts
were credit card holders and which were not. This was accomplished
via classification analysis using the See5 tool.
Finally, we examined all credit card holders to discern characterizations
of this particular sub-set of the bank's clientele. The Cobweb tool in
Weka was used to obtain the clusters used in this step.