ROC (blue) and precision-recall (orange)
Companion chart of the same model

Class Imbalance: ROC versus Precision-Recall Curves, Prevalence and Positive Predictive Value (2D)

See why a classifier with an excellent ROC curve can still raise mostly false alarms when the positive class is rare. Change class separation, prevalence and decision threshold and watch the ROC point stay put while precision collapses. A Blender-built data board with checks of AUC, Bayes precision and the F1 identity.

Made with 🟠 Blender