ROC vs PR Curve: Illusion of Optimism

Unbalanced Set of 1000 Data (100 Patients / 900 Healthy): Observe how ROC becomes unresponsive as false alarm (FP) increases, while PR collapses rapidly.

ROC Curve (Generally Optimistic)

PR Curve (Honest on Unbalanced Data)

0.50

Move the threshold to the left (e.g. to 0.20) to set off false alarms and watch the reaction of the PR graph.

Recall (Sensitivity / TPR) 0.0%
Precision 0.0%
FPR (False Alarm Rate) 0.0%
Confusion Matrix
Actual\Prediction Patient (Positive) Healthy (Negative)
Patient (100) 0TP (Yakalanan) 0FN (Missed)
Healthy (900) 0FP (False Alarm) 0TN (True Neg)
← Return to My Digital Space ← Apps