How do you evaluate anomaly detection?
mediumAnswer
- Usually severely imbalanced → don't use accuracy.
- Standard metrics: (1) ROC AUC (rank-based, threshold-free), (2) PR AUC (better under extreme imbalance), (3) precision@k and recall@k at operating threshold, (4) F1 for a chosen threshold.
- Cost-sensitive: business cost of false positive / false negative.
- Also: alert precision and MTTR (time to detect) in production ops contexts.
Check yourself — multiple choice
- Accuracy
- ROC AUC / PR AUC (better under imbalance) / P@k / R@k / cost-sensitive; alert precision + MTTR in production
- Random
- Only F1
AD evaluation: PR AUC, P@k, R@k, cost — not accuracy.
#anomaly-detection#evaluation
Practise Unsupervised Learning
214 interview questions in this topic.