Java Reference
In-Depth Information
javax::datamining
MiningObject
javax::datamining
BuildSettings
1
1
0..1
javax::datamining::base
ModelSignature
1
javax::datamining::base
Model
javax::datamining::base
ModelDetail
1
1
javax::datamining:attributeimportance
AttributeImportanceModel
javax::datamining::clustering
ClusteringModel
javax::datamining::association
A ssociation Model
javax::datamining::supervised
SupervisedModel
javax::datamining::supervised::classification
ClassificationModel
javax::datamining::supervised::regression
RegressionModel
Figure 8-10
Model objects and their relationships.
Figure 8-10 depicts the class diagram of the model-related
objects. Each model has an associated BuildSettings object. A model
may have an associated ModelDetail object. The ModelSignature
object provides details of the attributes needed to apply the model.
ModelSignature is defined as a separate object because it can be
used as input to validate the attributes of the datasets supplied to
the apply and test operations. For example, the decision tree model
discussed in Section 7.1 uses Age and Savings Account Balance as
predictors and hence these attributes are considered part of the
model signature. When applying this model, these attributes
should be provided. In addition, ModelSignature can be used to cre-
ate a default instance of PhysicalDataRecord that can be used for
record apply—as used for real-time scoring. Refer to PhysicalDa-
taRecordFactory interfaces in [JDM11 2006] for more details about
the PhysicalDataRecord create options.
8.1.4
Test Metrics
The quality of supervised models is evaluated by generating metrics
based on comparing predicted target values with actual target val-
ues. These metrics provide valuable insight into how a model per-
forms relative to test data, as well as how a model compares with
other models.
Search WWH ::




Custom Search