Phase 3 · Machine learning · Evaluation
ROC curves and AUC
Judge a classifier across every threshold.
Open this lesson Create a free account
Why it matters
Precision and recall depend on the threshold you picked. To compare models before anyone has chosen a threshold, you need a score that covers all of them at once. AUC is that score, and it's the one you've seen quoted all through this phase. It also has a plain meaning that carries over to ranking problems everywhere, from lead scoring to search results to judging an LLM grader.
What you'll be able to do
- Build a ROC curve from a classifier's scores across every threshold
- Interpret AUC as the chance a random positive is ranked above a random negative
- Use precision-recall curves when positives are rare
- Know what AUC doesn't tell you: the threshold, and whether probabilities are calibrated
Before this lesson
Next in this phase: Cross-validation