Classification Metrics and Recall Exam Question Explained
This AWS AI Practitioner question tests whether you can match the model metric to the business cost of each error. The key clue is that missing fraud is more expensive than investigating false alarms.
Short answer
The correct answer is B. Recall.
Recall is the correct answer because recall measures how many actual positive cases the model catches. In this scenario, the positive class is fraud. A false negative is a fraudulent transaction that the model fails to flag, and the question tells you that this is the expensive mistake. Optimizing recall pushes the model to catch a larger share of true fraud cases, even if that means more transactions get sent to investigation. That tradeoff is acceptable because false alarms are cheaper than missed fraud. Precision would answer a different question: when the model flags a transaction, how often is it truly fraud? Precision matters when false positives are the costly error, such as blocking a legitimate customer with no review path. Accuracy is a common trap because fraud datasets are usually imbalanced. A model can look accurate by predicting "not fraud" for nearly everything while still missing the rare fraud cases the business actually cares about. Specificity measures how well the model identifies true negatives, which again points away from catching fraud. On the exam, translate the scenario into false positives and false negatives before choosing the metric.
Practice Question
A fraud detection model flags suspicious transactions. The business cost of missing a fraudulent transaction is much higher than the cost of investigating a false alarm. Which metric should the team optimize for first?
Why B is correct
Recall is the correct answer because recall measures how many actual positive cases the model catches. In this scenario, the positive class is fraud. A false negative is a fraudulent transaction that the model fails to flag, and the question tells you that this is the expensive mistake. Optimizing recall pushes the model to catch a larger share of true fraud cases, even if that means more transactions get sent to investigation. That tradeoff is acceptable because false alarms are cheaper than missed fraud. Precision would answer a different question: when the model flags a transaction, how often is it truly fraud? Precision matters when false positives are the costly error, such as blocking a legitimate customer with no review path. Accuracy is a common trap because fraud datasets are usually imbalanced. A model can look accurate by predicting "not fraud" for nearly everything while still missing the rare fraud cases the business actually cares about. Specificity measures how well the model identifies true negatives, which again points away from catching fraud. On the exam, translate the scenario into false positives and false negatives before choosing the metric.
Why the other options are wrong
Precision reduces false positives. It is useful when a flagged transaction causes an expensive or irreversible action. Here false alarms are cheaper than missed fraud, so precision is not the first metric.
Accuracy can be misleading on imbalanced fraud data. If fraud is rare, a model can predict "not fraud" for almost every transaction and still have high accuracy while failing the business goal.
Specificity measures the true negative rate. It emphasizes correctly identifying non-fraud, but the question says the costly error is missing actual fraud.
The exam rule for precision vs recall
Precision and recall are both classification metrics, but they answer different business questions. Precision asks: of the cases the model predicted positive, how many were actually positive? It is true positives divided by true positives plus false positives. Use precision when false positives are expensive, disruptive, or dangerous. Examples include automatically suspending an account, denying a loan, or sending a patient into an invasive follow-up. Recall asks: of all the actual positive cases, how many did the model catch? It is true positives divided by true positives plus false negatives. Use recall when false negatives are expensive. Examples include missing fraud, missing disease, missing a safety defect, or failing to flag harmful content. Accuracy asks how often the model is right overall, but it hides error tradeoffs and often fails on imbalanced datasets. Specificity is the true negative rate and matters when correctly clearing negatives is the goal. For AIF-C01, do not memorize metrics as formulas only. Read the scenario and identify the more painful error. If missing the target event is worse, choose recall. If falsely flagging something is worse, choose precision. If both errors matter and the classes are imbalanced, F1 may be the best compromise.
Ready to see how you'd score?
Take a free 20-question diagnostic and find out which AWS Certified AI Practitioner domains you need to focus on. No signup required.
Practice 5 similar questions
Same cert, same or adjacent domain. Use these after reviewing the explanation.
Related AWS Certified AI Practitioner Practice Questions
Quick FAQ
What is the correct answer for this AWS Certified AI Practitioner question?
The correct answer is B. Recall. Recall is the correct answer because recall measures how many actual positive cases the model catches. In this scenario, the positive class is fraud. A false negative is a fraudulent transaction that the model fails to flag, and the question tells you that this is the expensive mistake. Optimizing recall pushes the model to catch a larger share of true fraud cases, even if that means more transactions get sent to investigation. That tradeoff is acceptable because false alarms are cheaper than missed fraud. Precision would answer a different question: when the model flags a transaction, how often is it truly fraud? Precision matters when false positives are the costly error, such as blocking a legitimate customer with no review path. Accuracy is a common trap because fraud datasets are usually imbalanced. A model can look accurate by predicting "not fraud" for nearly everything while still missing the rare fraud cases the business actually cares about. Specificity measures how well the model identifies true negatives, which again points away from catching fraud. On the exam, translate the scenario into false positives and false negatives before choosing the metric.
How should I study similar AWS Certified AI Practitioner questions?
Review the explanation, compare every distractor, then practice related questions in the same domain: Fundamentals of AI and ML.