AP Vs AUC-ROC
| Metric | Curve | Best For | Key Difference |
|---|---|---|---|
| Average Precision | Precision-Recall | Imbalanced data | Focuses on positive class |
| AUC-ROC | ROC (TPR vs FPR) | Balanced data | Considers both classes equally |
When to use AP:
- Highly imbalanced datasets (e.g., fraud detection, anomaly detection)
- When false positives and false negatives have different costs
- When you care primarily about the positive class
When to use AUC-ROC:
- Balanced datasets
- When both classes are equally important
- Standard binary classification