Question 9
Domain 3: Model DevelopmentWhen is ROC/AUC especially useful for classification evaluation?
Correct answer: A
Explanation
ROC/AUC is useful when you need a threshold-independent view of ranking performance because ROC plots true positive rate against false positive rate across all thresholds, and AUC summarizes that curve. This lets you compare classifiers by how well they rank positives above negatives without choosing a single cutoff.
Why each option is right or wrong
A. When you want a threshold-independent view of ranking performance
ROC/AUC is evaluated by sweeping the decision threshold and plotting true positive rate against false positive rate, so it measures how well the classifier ranks positives above negatives without committing to a single cutoff. In practice, that makes it the right choice when class probabilities or scores matter more than one fixed operating point, especially under class imbalance where accuracy at one threshold can be misleading.
B. When you need the average absolute error of a regression target
Average absolute error is MAE, a regression metric, not ROC/AUC.
C. When categories must be one-hot encoded
One-hot encoding is a categorical feature preprocessing step, not an evaluation metric.
D. When the model serves batch inference only
Batch inference is a deployment approach, not a classification evaluation criterion.