Why is a held-out TEST set required when evaluating a trained model — separate from the training and validation data?
A.It increases model accuracy during training
Test data is never used in training — it doesn't increase accuracy.
B.It measures performance on truly unseen data, detecting overfitting
Unseen-data evaluation = the test set's purpose. Correct.
C.It tunes hyperparameters
Hyperparameter tuning uses the VALIDATION split.
D.It reduces training cost
Test sets don't reduce cost — they're about honest evaluation.
Why this answer
The test set stays untouched until final evaluation — it estimates real-world performance on data the model never saw. The VALIDATION set guides tuning; if you tune to the test set it leaks and stops being an honest measure.
Correct answer: B · Last reviewed 2026-11-05
Official AWS references
Practice this interactively
Answer without seeing the solution first in the AI/ML Foundations practice session.
Practice AI/ML Foundations