JDM apply settings were designed to provide flexibility in defining
the results from model apply. Section 3.3.2 introduced the model
apply operation. In general, all supervised models and some unsu-
pervised models such as clustering can be applied to produce apply
results—a table or view in a database or a file in a file system.
Apply settings specify the desired contents in the apply output data .
The specific content varies by mining function, hence JDM defines
function-specific apply settings interfaces. Figure 8-7 depicts various
apply settings interfaces and their relationships. The ApplySettings is
the base interface that defines the common settings across all function-
level apply settings. ClassificationApplySettings, RegressionApplySettings,
and ClusteringApplySettings interfaces specify the settings for applying
classification, regression, and clustering models, respectively.
Apply for a classification model computes the target value and
corresponding probability for each case of the specified apply input
data . Normally, the target value with highest probability is consid-
ered the top prediction. However, the probability is not always the
best measure to find the top prediction. If the cost matrix is specified
for model build or apply, the cost values are computed in addition to
probability. We may want to select the target value that has the least
cost as the top prediction.
For example, in the customer attrition problem discussed in Sec-
tion 7.1, the target attribute attrite can have one of two possible tar-
get values Attriter or Non-Attriter . Note that the model cannot
predict attrition with 100 percent probability for each case. In this
example, let us say the model predicted that a customer is 30 per-
cent likely to attrite, meaning 70 percent unlikely to attrite. As such,
the value Non-attriter is selected as the top prediction since it has the
Apply settings objects.