DataTerminology.comNEW
Also known as: AUROC
Area under the ROC curve, measuring a classifier’s ability to distinguish classes.
DESCRIPTION
AUC (Area Under the ROC Curve) summarizes a classifier’s ranking quality across thresholds. It measures the probability a randomly chosen positive is ranked higher than a negative. AUC is threshold-independent, enabling broad comparison. It is robust to class imbalance for ranking but not for probability calibration. Complementary metrics like PR-AUC may be preferred for highly imbalanced positives.
ALTERNATIVE DEFINITIONS
scikit-learn
Compute Area Under the Receiver Operating Characteristic Curve (ROC AUC) from prediction scores.
Defines the metric as implemented in a widely used library.
Wikipedia
The area under the ROC curve (AUC) is equal to the probability that a classifier will rank a randomly chosen positive instance higher than a randomly chosen negative one.
Highlights the probabilistic interpretation of AUC.
scikit-learn
AUC provides an aggregate measure of performance across all classification thresholds.
Emphasizes threshold-independent evaluation.
RELATED TERMS
ROC Curve TPR False Positive Rate Threshold Binary Classification Calibration Confusion Matrix Lift F1 Score
COMPETING TERMS
RELATIONS
- Broader: Score
KNOWLEDGE AREAS
Mathematics Data Science Testing & Validation AI
TARGET AUDIENCE
Data Scientist AI Engineer MLOps Engineer BI Specialist
EXTERNAL SOURCES
- Wikipedia - Receiver operating characteristic
- scikit-learn - User guide for ROC/AUC
- Google Developers - MLCC ROC/AUC primer
TOP QUESTIONS
When should I prefer AUC over accuracy?
AUC is preferable when thresholds vary or classes are imbalanced. It evaluates ranking performance across all possible thresholds.
Category: Guidance
What is a good AUC Score?
AUC of 0.5 indicates random ranking, while 1.0 is perfect. Many practical models range 0.7–0.9 depending on difficulty.
Category: Opinion
Is AUC affected by class imbalance?
AUC is relatively robust for ranking but can be misleading when positive prevalence is extremely low. PR-AUC may better reflect rare events.
Category: Risks
How is AUC computed?
It integrates the ROC curve or uses pairwise ranking relationships. Implementations often apply the trapezoidal rule.
Category: Example
Does higher AUC guarantee better calibration?
No. AUC measures ranking, not probability calibration. Use calibration metrics like Brier score or reliability plots.
Category: Misunderstanding
What are common mistakes using AUC?
Comparing AUC across very different datasets or ignoring confidence intervals are common mistakes. Also, conflating AUC with PR-AUC.
Category: Mistakes
How do ties affect AUC?
Ties in scores reduce AUC slightly, since some positive-negative pairs are indistinguishable. Implementations typically handle ties consistently.
Category: Example
Can AUC be extended beyond binary classification?
Yes, via one-vs-rest or macro/micro averaging strategies. Interpretability becomes more complex with many classes.
Category: Guidance
What is the relationship between AUC and Wilcoxon-Mann-Whitney?
AUC equals the Wilcoxon-Mann-Whitney statistic normalized by the number of positive-negative pairs. Both measure rank separation.
Category: Example