Databases Reference
In-Depth Information
Figure 5.11
ROC Curve - ANN Target
curve
(AUC). Given that FPR and TPR both range between 0 and 1, the AUC of
the ideal classifier is 1.0. The dotted line in the ROC viewer represents the ROC
of a random guess. Its AUC is 0.5. Hence, for a classifier to be considered useful
in any way, its AUC should be greater than 0.5.
Indicate to the ROC viewer which class value you define as the positive
result by selecting “Y” in the “Positive result” drop-down.
The ROC viewer presents two curves - one computed using the training
data and the other using the validation data. The AUC of both curves is high
(above 0.96 in Figure 5.11). Given that the classification modeler did not see
the validation data until after the model was complete, the validation ROC
curve should normally be used when judging the performance of classifiers
and in making comparisons between classifiers.
