Using the JDM API - Java Data Mining: Strategy, Standard, and Practice - page 234

Table 9-18 Attribute importance-related interfaces

attributeimportance.AttributeImportanceSettings extends base.BuildSettings

Methods Description

set/getMaxAttributeCount(int maxCount) Used to set/get the maximum number of attributes to be

included in the resulting model.

set/getTargetAttributeName(String) Used to set/get the target attribute for supervised

attribute importance.

isSupervised():boolean Returns true if a target attribute has been set.

attributeimportance.AttributeImportanceModel extends base.Model

Methods Description

getAttributesByPercentage(double percent,
SortOrder ordering):Collection

Returns a collection of attributes whose rank is within
the specified percentage of the entire attributes.

getAttributesByRank(int lowerRank, int
upperRank):Collection

Returns a collection of attributes whose rank is within
the specified range.

getAttributesByRank(SortOrder ordering):
Collection

Returns a collection of all attributes in the specified
order.

getMaxRank():int Returns the rank number of the least important

attribute.

getAttributeCount():int Returns the number of attributes ranked in the model.

Attribute Importance Settings. set Target AttributeName(String) , the DME
will build a supervised attribute importance model. In this case, the
model ranks the attributes according to their ability to predict the
specified target attribute. The DME will build an unsupervised
model if a target attribute is not specified. In this case, the model
ranks attributes according to their contribution to the desired unsu-
pervised model.
Listing 9-15 shows the code that builds an attribute importance
model for the CUSTOMERS dataset to find the importance of predic-
tor attributes used for predicting the attrite target attribute. Lines 28
and 29 show the creation of an AttributeImportanceSettings object and
sets the target attribute to attrite .
In some domains, data may have thousands of attributes. For
example, in gene analysis, thousands of gene types can be collected to
find their effect on a particular disease; each type can be included as
an attribute in the dataset. To reduce the size of the attribute impor-
tance model by filtering the attributes at the time of model building,