Java Reference

In-Depth Information

Transformed

Dataset using

same build data

transformations

and statistics

Apply

Settings

New Dataset

(with
unknown

target values)

Transform,

Prepare

Data

Apply

Model

Apply

Result

Apply Data

Apply Data
´

Model

Figure 3-8

Data mining model apply process.

As illustrated in Figure 3-8, we begin with a new dataset that we

wish to apply the model to—the
apply data
. The apply data must be

transformed using the same transformations as for the build data.

This transformed dataset is then used with the model and
apply set-

tings
to produce the
apply result
. The apply settings describe the con-

tents that the user wants in the apply results (e.g., the top

prediction(s) for a case, the probability that the prediction is correct,

additional attributes carried over from the apply data, and so on).

These are explained further in Chapters 8 and 9.

The apply result is typically a table where each input case from

the apply data has a corresponding output case. A unique identifier

of the case is normally provided in the apply result so that results can

be matched to the apply data. For example, you likely want to know

which customer is predicted to respond favorably to a campaign.

3.3.3

Model Test

Model test applies only to supervised models—classification and

regression. The reason for this is that in order to test a model, you

need to know the
correct
outcome to determine how accurate the

model is. In unsupervised models, we do not have a target (known

outcome) and so there is no known value to compare. In general, we

Search WWH ::

Custom Search