A team has trained multiple models for a fraud detection system. The business has two requirements: (1) minimize false negatives (missed fraud) and (2) keep the false positive rate below 5%. Model A has a recall of 0.95 and a false positive rate of 8%. Model B has a recall of 0.90 and a false positive rate of 4%. Model C has a recall of 0.85 and a false positive rate of 2%. Which model should they select?

Model B, because it meets the false positive rate constraint and has the highest recall among the models that meet the constraint.

Combine models A and B to get both high recall and low false positive rate.

Model C, because it has the lowest false positive rate.

Model A, because it has the highest recall, which is the most important metric.

During training of a deep learning model, the loss curve shows high variance and the model is not converging. The data scientist suspects the learning rate is too high. What should they do to address this?

Increase the batch size to reduce variance, but keep the learning rate the same.

Switch to a different optimizer, such as Adam, without changing the learning rate.

Decrease the learning rate and retrain from scratch.

Implement a learning rate schedule, such as reducing the learning rate after a plateau or using a cosine decay.

During training of a neural network, the data scientist observes that the training loss decreases steadily, but the validation loss starts to increase after epoch 10. The team needs a model that generalizes well. What should they do?

Implement early stopping with a patience of 5 epochs and monitor validation loss, while also considering regularization techniques.

Stop training immediately and use the model from epoch 10 as the final model.

Ignore the validation loss and continue training until the training loss converges.

Increase the learning rate to help the model escape the overfitting regime.

What is the purpose of model selection?

To choose the best model based on validation metrics and business criteria

Which metric is commonly monitored during training to detect overfitting?

Difference between training and validation loss

PMICertified Professional in Managing AI (PMI-CPMAI)

Domain 4Objective 3

Manage AI/ML Model Training PMI-CPMAI Practice Questions (Page 7)

Part of the Manage AI Model Development and Evaluation domain, which accounts for 16% of the PMI-CPMAI exam. PMI does not publish an official question count, but from its 160-minute exam (~65–105 total, ~10–17 in this domain), expect 2–3 from this objective — we provide 36 practice questions to prepare you well beyond it. (estimate)

36questions here

8free pages

10concepts

16%of the exam

Questions 31–35

31expert · hard
A team has trained multiple models for a fraud detection system. The business has two requirements: (1) minimize false negatives (missed fraud) and (2) keep the false positive rate below 5%. Model A has a recall of 0.95 and a false positive rate of 8%. Model B has a recall of 0.90 and a false positive rate of 4%. Model C has a recall of 0.85 and a false positive rate of 2%. Which model should they select?
Select an answer first
32application · medium
During training of a deep learning model, the loss curve shows high variance and the model is not converging. The data scientist suspects the learning rate is too high. What should they do to address this?
Select an answer first
33application · medium
During training of a neural network, the data scientist observes that the training loss decreases steadily, but the validation loss starts to increase after epoch 10. The team needs a model that generalizes well. What should they do?
Select an answer first
34foundation · easy
What is the purpose of model selection?
Select an answer first
35foundation · easy
Which metric is commonly monitored during training to detect overfitting?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

Next 5 questions →Back to PMI-CPMAI

Free Basic Practice is a study aid with revealable answers — not a scored exam. Examers.io is independent and not affiliated with or endorsed by PMI. “PMI-CPMAI” is a trademark of its owner, used for identification only.