Databases Reference
In-Depth Information
Figure 5.12
Lift Chart
of the validation observations (green) are positive. Thus, if we only select the
top 20%, we would have a false positive rate of 1.4% on the training observa-
tions and 3.7% on the validation observations.
Gradually move to the right along the X axis. Note the changes in selected
positive training and validation percentages. What is the greatest “Selected
Cutoff” percentage that you can achieve while maintaining a “100.0%
positive” rate for the validation data?
The horizontal gray bar represents the total number of positive observations
in the training set. As the red line approaches that bar, it becomes more and more
difficult to select additional positive observations. Once all positive training
observations have been selected, the red line flattens. The same is true for the
green validation line.
As noted previously, the costs associated with false positives and the costs
associated with false negatives in a classifier's application are usually different.
If those costs can be associated with a monetary amount, then resulting
monetary gains/losses can be plotted.
Select the “$” button. A dialog box opens allowing you to specify monetary
parameters that are applied to a simple cost/benefit model.
For the Wheels example, enter the following:
