MLA-C01 Practice Question 75: Precision Recall Calculation
AWS Certified Machine Learning Engineer – Associate (MLA-C01) · Test 8 · question 5 of 10
A fraud classifier is evaluated on 10,000 validation transactions, 200 of which are actually fraudulent. The results are 80 true positives, 20 false positives, 120 false negatives, and 9,780 true negatives. The fraud team asks two questions: when the model flags a transaction, how often is it really fraud, and what share of all fraud does the model catch?
Which values answer those two questions?
Show answer and explanation
Answer:
*When the model flags a transaction, how often is it fraud* is precision: true positives divided by all positive predictions, TP / (TP + FP) = 80 / (80 + 20) = 0.80. *What share of all fraud does it catch* is recall: true positives divided by all actual positives, TP / (TP + FN) = 80 / (80 + 120) = 0.40. The model is usually right when it flags something but misses most of the fraud (option A).
- B is wrong because it swaps the two metrics. Precision uses false positives in its denominator, and recall uses false negatives. - C is wrong because 0.998 is the true negative rate (specificity), TN / (TN + FP) = 9,780 / 9,800. That measures how well legitimate transactions are cleared, not how much fraud is caught. - D is wrong because 0.986 is overall accuracy, (TP + TN) / total = 9,860 / 10,000. The 9,780 true negatives inflate it, which is why accuracy misleads on imbalanced data.
Reference: Metrics and validation