Module 1 Assessment: Algorithm Selection Readiness

Estimated active time: 30-45 minutes

Question 1

A model with the highest validation score is also slow, hard to explain, and only slightly better than a simple baseline. What should you check before choosing it?

Pass answer: mention simpler baseline comparison, decision cost, runtime or maintenance, error pattern, and whether the gain is meaningful.

Question 2

A dataset has 300 rows, 80 features, missing values, and a noisy label. Why might a very flexible model be risky?

Pass answer: mention overfitting, instability, noise learning, validation uncertainty, and the need for a simple baseline.

Question 3

Why should the test set not be used repeatedly while choosing algorithms?

Pass answer: repeated use can tune choices to the test set and make the final result too optimistic.

Question 4

Give one reason to reject an algorithm before fitting it.

Pass answer: any defensible reason such as feature mismatch, scale sensitivity without preprocessing, runtime, interpretability need, data size, missing values, metric mismatch, or unsupported prediction-time features.

Question 5

What is a baseline ladder?

Pass answer: a planned sequence from dummy baseline to simple useful model to interpretable candidate to more complex candidate, where each higher step must justify its extra complexity.