AIF-C01 Practice Question: Deploying a machine learning model to detect…

A company is deploying a machine learning model to detect fraudulent transactions. The dataset is highly imbalanced (1% fraud). The team needs to evaluate model performance and minimize false positives while maintaining high recall. Which TWO metrics should they focus on? (Select TWO.)

⚠ Common exam trap

A common misconception is that accuracy is a reliable metric for imbalanced datasets, or that F1 score alone is sufficient when the question explicitly asks for two separate metrics to independently control false positives and recall.

Answer choices

Why each option matters

Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.

Correct answer & explanation

✓

Recall

Option B (Recall) is correct because with only 1% fraud, recall measures the proportion of actual fraudulent transactions the model successfully catches, which directly supports the requirement to maintain high recall and avoid missing fraud cases. Option C (Precision) is correct because minimizing false positives means reducing legitimate transactions incorrectly flagged as fraud, and precision is exactly the ratio of true positives to all positive predictions (TP / (TP + FP)), so higher precision directly reflects fewer false alarms. Accuracy (A) is misleading on a 1% imbalanced dataset, since a model predicting 'no fraud' for everything would score 99% accuracy while catching zero fraud. F1 score (D) balances precision and recall but does not by itself target the specific goal of minimizing false positives while keeping recall high. AUC-ROC (E) summarizes ranking performance across thresholds but does not directly express the false-positive/recall trade-off the team must optimize.

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

✗
Accuracy
Why it's wrong here
Accuracy is dominated by the 99% legitimate class, so a model predicting "no fraud" always scores 99% while catching zero fraud — it cannot expose false positives or recall. It is tempting because accuracy suits balanced datasets where classes contribute equally, but here it masks the minority-class failures the team must minimise.
✓
Recall
Why this is correct
Recall measures the proportion of actual frauds correctly identified, directly addressing the requirement to maintain high recall on the 1% minority class. It exposes missed frauds (false negatives), which are the costly errors when the positive class is rare.
✓
Precision
Why this is correct
Precision directly quantifies the proportion of predicted frauds that are genuinely fraudulent, so it penalises false positives — the stem's explicit constraint. Because the 1% fraud rate makes accuracy misleading, precision paired with recall gives a clearer picture of performance on the minority class.
✗
F1 score
Why it's wrong here
F1 score collapses precision and recall into one harmonic mean, so it cannot expose the false-positive rate the team must minimise; a model can hold a respectable F1 while flagging many legitimate transactions. It suits balanced datasets where a single trade-off summary is wanted, not this imbalanced fraud case.
✗
AUC-ROC
Why it's wrong here
AUC-ROC aggregates performance across all thresholds and stays optimistic under severe class imbalance, because the false-positive rate is diluted by the large legitimate class. It is tempting for threshold-independent model comparison, but the team needs precision and recall at the operating threshold to control false positives while maintaining recall.

Go deeper

AIF-C01 Practice Question: Deploying a machine learning model to detect…

Why each option matters

Option-by-option breakdown

Related to this question