Java Data Mining Concepts - Java Data Mining: Strategy, Standard, and Practice - page 156

mining function such as minimum description length [Wikipedia 2005]
and even decision tree . Decision trees inherently select the best
attributes for branching in the tree and normally include far fewer
attributes in the resulting model than originally input.
JDM allows a model's signature to contain attribute importance
rankings that result as a by-product of model building. After a model
is built, the user can determine the relative importance of the
attributes as determined by the specific algorithm. Since different
independent attribute importance algorithms can produce signifi-
cantly different results, having access to the importance assigned by
the algorithm is more accurate.

7.3.4 Use Model Details: Explore Attribute Importance Values

An attribute importance model produces relative importance values
by which attributes are ranked. In this example, Table 7-9 lists the

Table 7-9 Attribute importance values

Attribute name Importance Rank

Education 0.46 1

Occupation 0.40 2

Age 0.36 3

Marital Status 0.34 4

Avg. Savings Balance 0.31 5

Home loan balance 0.27 6

Annual Income 0.22 7

Retirement Balance 0.16 8

Avg. Checking Balance 0.10 9

Capital Gain 0.09 10

City 0.06 11

Ethnic Group 0.02 12

Native Country 0.03 13

County 0.06 14

State 0.11 15