Concept - ROC Curve and AUC

• ROC (Receiver Operation Characteristic): illustrates the diagnostic ability of a binary classifier as its discrimination threshold varies; plots TPR against FPR at various thresholds (s.281). Plotted parametrically as TPR(T) vs FPR(T) with threshold T varying (s.283).
• $FPR=\frac{FP}{FP+TN}$ (probability of false alarm); $TPR=\frac{TP}{TP+FN}$ (probability of detection) (s.282).
• Best possible: upper-left corner (0,1) = perfect classification; a random guess lies on the diagonal; above the diagonal = better, below = worse (s.281-282).
• AUC (Area Under Curve): area under the ROC curve. Larger is better: AUC(A) > AUC(B) means classifier A is better than B (s.283). AUC = probability the classifier ranks a randomly chosen positive example higher than a randomly chosen negative one; one of the most widely used metrics for binary classification (s.283).

Related

Concept - Confusion Matrix | Concept - Precision and Recall | Concept - Logistic Regression

Source chunk: 03 Text Classification - Chunk 10 (s.271-300)