A team is building a model to predict whether a patient has a specific disease. They have a dataset of 5,000 patients, each with a confirmed diagnosis. They split the data into 70% training and 30% test sets. After training, they evaluate the model on the test set and get high accuracy. However, they are concerned about the model's performance on different patient demographics. Which approach best addresses this concern?

Use a random train-test split without stratification

Use only accuracy as the metric to evaluate the model

Train the model on the entire dataset without a test set

Use stratified cross-validation to ensure each fold has a similar distribution of demographics and disease outcomes

A team is training a gradient-boosted decision tree model on a dataset with 50,000 rows and 200 features. The model achieves 99.8% accuracy on the training set but only 82% accuracy on a held-out test set. The team needs to improve generalization without losing the ability to capture complex nonlinear relationships. Which approach is most appropriate?

Increase the number of boosting rounds and add more features

Apply stronger regularization (e.g., higher L2 penalty) and use cross-validation to tune hyperparameters

Reduce the training set size to force the model to learn simpler patterns

Switch to a linear model to eliminate overfitting entirely

Why is it important that the testing dataset is not used during the training process?

To reduce the amount of labeled data needed for training

To avoid overfitting and obtain an unbiased estimate of model performance

To ensure the model is trained on a smaller dataset for faster convergence

To allow the model to learn from a wider variety of examples

A data scientist trains a support vector machine with a high-degree polynomial kernel on a small dataset. The model performs perfectly on training data but poorly on validation data. Which action is most likely to improve validation performance?

Reduce the polynomial degree or use a linear kernel

Add more features without changing the kernel

GIAC Machine Learning Engineer

Domain 2Objective 3

Supervised Learning GMLE Practice Questions (Page 6)

Part of the Machine Learning Algorithms domain, which makes up ~29% of our current practice bank. GIAC (SANS) does not publish an official question count, but from its 180-minute exam (~70–120 total, ~20–35 in this domain), expect 7–12 from this objective — we provide 40 practice questions to prepare you well beyond it. (estimate)

40questions here

8free pages

6concepts

Questions 26–30

26application · medium
A team is building a model to predict whether a patient has a specific disease. They have a dataset of 5,000 patients, each with a confirmed diagnosis. They split the data into 70% training and 30% test sets. After training, they evaluate the model on the test set and get high accuracy. However, they are concerned about the model's performance on different patient demographics. Which approach best addresses this concern?
Select an answer first
27foundation · easy
For a binary classification model, which metric is most appropriate when the cost of false negatives is very high (e.g., missing a cancer diagnosis)?
Select an answer first
28expert · hard
A team is training a gradient-boosted decision tree model on a dataset with 50,000 rows and 200 features. The model achieves 99.8% accuracy on the training set but only 82% accuracy on a held-out test set. The team needs to improve generalization without losing the ability to capture complex nonlinear relationships. Which approach is most appropriate?
Select an answer first
29foundation · easy
Why is it important that the testing dataset is not used during the training process?
Select an answer first
30application · medium
A data scientist trains a support vector machine with a high-degree polynomial kernel on a small dataset. The model performs perfectly on training data but poorly on validation data. Which action is most likely to improve validation performance?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

Next 5 questions →Back to GMLE

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 GIAC (SANS). “GMLE” is a trademark of its owner, used for identification only.

Supervised Learning GMLE Practice Questions (Page 6)

Questions 26–30

For a binary classification model, which metric is most appropriate when the cost of false negatives is very high (e.g., missing a cancer diagnosis)?

Why is it important that the testing dataset is not used during the training process?

A data scientist trains a support vector machine with a high-degree polynomial kernel on a small dataset. The model performs perfectly on training data but poorly on validation data. Which action is most likely to improve validation performance?

Keep studying — Machine Learning Algorithms