Graded Quiz: Evaluating and Validating Machine Learning Models :Machine Learning with Python (IBM AI Engineering Professional Certificate) Answers 2025

1. Question 1

In medical diagnosis, which metric is most important?

• ✅ Recall

• ❌ Accuracy

• ❌ F1 Score

• ❌ Precision

Explanation:
Recall measures how many actual positive cases were correctly detected.
In healthcare, missing a true positive can be dangerous, so recall is critical.

2. Question 2

Which regression metric is the square root of MSE?

• ❌ R-squared

• ❌ MAE

• ❌ MAE

• ✅ Root Mean Squared Error (RMSE)

Explanation:
RMSE = √MSE, giving error in the same unit as the target variable.

3. Question 3

Best metric to evaluate cluster separation?

• ✅ Silhouette score

• ❌ Elbow method

• ❌ Evaluation method

• ❌ Davies-Bouldin Index

Explanation:
Silhouette score measures cohesion vs separation.
Higher = better-separated clusters.

4. Question 4

Model performs well on training but poorly on test data → problem?

• ✅ Overfitting

• ❌ Cross-validation

• ❌ Train-test split

• ❌ Data snooping

Explanation:
Overfitting = model memorizes training data but fails to generalize.

5. Question 5

Difference between Lasso and Ridge?

• ❌ Lasso only for feature selection

• ✅ Lasso uses L1 penalty, Ridge uses L2 penalty

• ❌ Lasso uses larger datasets

• ❌ Ridge uses L1

Explanation:
Lasso (L1) → can shrink coefficients to zero (feature selection).
Ridge (L2) → shrinks coefficients but not to zero.

6. Question 6

How to mitigate data leakage?

• ❌ Use single feature

• ✅ Avoid including features derived from the entire dataset

• ❌ Shuffle data

• ❌ Ensure proper splits only

Explanation:
If a feature is computed using global dataset knowledge (e.g., overall average), it leaks test information into training.

7. Question 7

Interpreting feature importance without checking relationships leads to:

• ❌ Missing scaling

• ❌ Using minimal-target features

• ❌ Using uncorrelated features only

• ✅ Overlooking correlated features in importance scores

Explanation:
If features are correlated, importance scores can be misleading (importance gets split among correlated features).

🧾 Summary Table