Home > Testing and Tuning Models > Testing Classification Models > Test Metrics for Classifica... > Classification Model Test a... > Tuning Classification Models > Cost > Lift
Lift measures the degree to which the predictions of a classification model are better than randomly-generated predictions. For more information, see Lift.
Follow these steps to tune a model using lift:
Before you tune the model, you must select Generate Select Test Results for Model Tuning and run the node, as described in Tuning Classification Models.
Open Properties: for the build node: right-click the node and select Go to Properties from the context menu.
Select the models that you want to tune and click the tune icon in the menu bar.
Select Tune from the drop down menu. The Tune Settings dialog opens in a new tab. In Tune Settings, go to the Lift tab.
If you are tuning more than one model, select a model from the Models list in the bottom pane. After you tune the first model, you return to this pane and select another model.
Select the target value for tuning from the Target Value list.
Decide whether to tune using the Cumulative Positive Cases chart, the default or the Cumulative Lift chart. Select the chart from the Display list.
Either chart displays several curves: the lift curve for the model that you are tuning, ideal lift, and Random lift, the lift from a model where predictions are random.
The chart also displays a blue vertical line that indicates Threshold, the quantile of interest.
Selected a quantile using the slider in the quantile display below the lift chart. As you move the slider, the blue vertical bar moves to that quantile, and the tuning panel is updated with the Performance Matrix for that point.
Click Tune, below the Performance Matrix. New Tune Settings are displayed in the same panel as the Performance Matrix.
Examine the Derived Cost Matrix. You can continue tuning by changing any selections that you made.
To cancel the tuning, click Reset. Tuning returns to Automatic.
When you done tuning, either click OK to accept the tuning or Cancel to cancel.
To see the impact of the tuning, run the model node.