← Confusion Matrix All techniques Next: Tokenization →

29 — ROC Curve & AUC

Walk the threshold from 1.0 down to 0.0 and plot TPR vs FPR at each step. The curve summarizes performance at every cutoff. AUC = area under it; 1.0 = perfect, 0.5 = random.

AUC
Gini = 2·AUC − 1
TPR at threshold
FPR at threshold
Score distribution (positives vs negatives)
ROC curve (current threshold = orange dot)
0.50
How to read it: a great model hugs the top-left corner — high TPR at low FPR across all thresholds. The diagonal is the "coin flip" baseline (AUC = 0.5). Use AUC to compare models without committing to a threshold; use the threshold slider when you need to pick one.