AIF-C01 Practice Question 81: Underfitting High Error Both Sets
AWS Certified AI Practitioner (AIF-C01) · Test 9 · question 1 of 10
A data science team trains a model to predict delivery times. It scores 61% accuracy on the training data and 60% on the held-out test data. The two scores are almost identical, but both are far below the business target of 90%. What does this pattern most likely indicate?
Show answer and explanation
Answer:
AWS describes underfit models this way: *"Underfit models exhibit high bias, meaning they produce inaccurate results on both the training data and the test set."* Poor scores on the training data itself, with almost no gap to the test score, mean the model has not learned the underlying pattern. Typical remedies are a more expressive model, more informative features, or more training time (option B).
- A is wrong because overfitting shows the opposite signature: very good training scores and much worse test scores. More regularization would reduce capacity further and make underfitting worse. - C is wrong because leakage of test data into training makes scores look better than reality. Here both scores are poor. - D is wrong because a small train/test gap only shows that the model behaves consistently. It does not show that the model is good; here it fails the 90% target on data it has already seen.
Reference: What is Overfitting?