ROC AUC (area under the ROC curve)

The ROC curve plots a classification model's true positive rate against its false positive rate as the decision threshold is varied from strict to lenient. ROC AUC is the area under that curve, a single number from 0 to 1 that summarizes performance across every possible threshold at once, rather than at one fixed cutoff the way a confusion matrix does.

An AUC of 0.5 means the model performs no better than random guessing; 1.0 means perfect separation between classes. Intuitively, AUC is the probability that the model ranks a randomly chosen positive example higher than a randomly chosen negative one, which is why it is popular for comparing models or evaluating a model's ranking ability before a specific threshold has been chosen for deployment.

AUC's main limitation is that it treats all thresholds and both error types as equally important, which is rarely true in a real decision, and it can look misleadingly high on strongly imbalanced datasets where a model does well simply by favoring the majority class. In those cases, practitioners often pair AUC with precision and recall or the f1 score at the specific threshold the model will actually operate at, rather than relying on AUC alone.