ROC Curve & AUC

Question 1

The ROC curve plots…

Accepted Answer

True Positive Rate vs False Positive Rate — The Receiver Operating Characteristic curve traces TPR (recall) against FPR as the decision threshold is varied from 0 to 1.

Question 2

AUC stands for Area Under the (ROC) Curve and ranges from…

Accepted Answer

0 to 1, where 0.5 is random — AUC summarizes the ROC into one number: 0.5 means no better than chance, 1.0 means perfect ranking of positives above negatives.

Question 3

A key property of AUC is that it is…

Accepted Answer

Threshold-independent, evaluating ranking across all cutoffs — AUC evaluates how well the model ranks positives above negatives regardless of any single cutoff, so you needn't fix a threshold to compare models.

Question 4

AUC has the probabilistic interpretation:

Accepted Answer

The probability a random positive ranks higher than a random negative — AUC = P(score(positive) > score(negative)) for a random positive/negative pair — a clean measure of discrimination.

Question 5

On the ROC plot, a random classifier produces…

Accepted Answer

The diagonal line from (0,0) to (1,1) — Random guessing gives TPR ≈ FPR at every threshold, tracing the diagonal (AUC = 0.5).

Question 6

The ideal ROC operating point is near…

Accepted Answer

Near the top-left corner of the plot — Top-left means catching almost all positives (TPR→1) while raising almost no false alarms (FPR→0) — the best trade-off.

Question 7

On heavily IMBALANCED data, ROC-AUC can be misleading, so you should also consider…

Accepted Answer

The Precision-Recall curve as well — With few positives, FPR stays low even with many false positives, flattering ROC; PR curves better reflect minority-class performance.

Question 8

The y-axis of the ROC curve, True Positive Rate, equals…

Accepted Answer

Recall, computed as TP over TP plus FN — TPR is recall/sensitivity — the fraction of actual positives correctly identified.

Question 9

The x-axis of the ROC curve, False Positive Rate, equals…

Accepted Answer

FP over FP plus TN, or one minus specificity — FPR is the fraction of actual negatives wrongly flagged positive; it is one minus specificity.

Question 10

As you LOWER the classification threshold, you move along the ROC curve toward…

Accepted Answer

Toward the upper-right area of the plot — A lower threshold predicts more positives, raising both true and false positive rates — sliding up-right along the curve.

Data Parameters

Decision Threshold

Actions

ROC Curve

Metrics at Threshold = 0.50

Predicted Probability Distributions

ROC or precision-recall?

Test your knowledge

The ROC curve plots…

Learn these first

Interview Questions

References & Further Reading

Love this visualization?