AUC
AUC is the area under the ROC curve, equivalent to the probability that a random positive sample scores higher than a negative one, measuring a binary classifier's overall discrimination.
Purpose
To provide a threshold-independent binary classification metric that is robust to class imbalance.
Mechanism
AUC equals the integral of the ROC curve, i.e., P(score(positive) > score(negative)). It is computed by the trapezoidal rule or the Mann-Whitney U statistic. 0.5 is random, 1.0 is perfect, and 0 is fully inverted.
Use cases
- Classifier comparison
- Imbalanced evaluation
- Model selection