ROC-AUC

In one line

The area under the ROC curve — a threshold-independent measure of how well a classifier ranks positives above negatives.

What it actually means

The ROC curve plots true-positive rate against false-positive rate as you sweep the decision threshold from 1 to 0. AUC is the integral under that curve. It has a clean probabilistic interpretation: AUC is the probability that a randomly chosen positive scores higher than a randomly chosen negative. Random guessing gives 0.5, perfect ranking gives 1.0. Because it considers all thresholds, AUC tells you about the quality of the ranker itself, not a specific operating point.

Why it matters

AUC is the right metric when you care about ranking — credit risk scoring, ad CTR, search relevance — and when downstream consumers may pick their own threshold. It’s the wrong metric when classes are very imbalanced and you only care about the top of the list (use PR-AUC or precision@k there) or when calibration matters more than ordering.

Example

from sklearn.metrics import roc_auc_score
auc = roc_auc_score(y_true, y_score)  # y_score = predicted probabilities or any monotone function of them

You’ll hear it when

• Building a credit, fraud, or risk model.
• Comparing classifiers without committing to a threshold.
• Reading a Kaggle competition writeup.
• Discussing PR-AUC vs ROC-AUC on imbalanced data.
• Validating that a binary classifier is at least better than random.