Multi-choice: Classification

Question 1

Q1 (Classification). Maximum number of support vectors in a two-class dataset of N points in a hard SVM? (One choice) 1) 2
2) N
3) N-2
4) N+2
5) N/2
6) 2N

Answer 1

Correct item: 2. Explanation: In the worst case, all points lie on the margin, becoming support vectors.

Question 2

Q2 (Classification). Consequences of increasing K in K-nearest neighbor? (Three correct)
1) Overfitting large
2) Smoother boundary
3) Underfitting more possible
4) Classes with few data may disappear
5) Classes with many data may disappear
6) Faster training

Answer 2

Correct items: 2, 3, 4. Explanation: Higher K => smoother boundary, more underfitting, and small classes can vanish in majority voting.

Question 3

Q3 (Classification). Bayesian Logistic Regression approximate solutions. Which is correct? (One choice)
1) No relation between MAP and Laplace approx.
2) Laplace approx. needs sampling
3) Laplace approx. is exact if posterior is Gaussian

Answer 3

Correct item: 3. Explanation: Laplace approximation is indeed exact if the posterior is exactly Gaussian.

Question 4

Q4 (Classification). About performance metrics. (One choice) 1) Specificity & sensitivity are independent
2) If TP rises, TN always rises
3) We want high TP and TN
4) We want high TP and FP

Answer 4

Correct item: 3. Explanation: We aim to maximize true positives and true negatives.

Question 5

Q5 (Classification). What is true about Bayes classifier? (One choice)
1) Training is faster than k-NN
2) Same decision boundary as SVM
3) Naive Bayes has a faster training phase than k-NN
4) Naive Bayes classifier makes its training more efficient

Answer 5

Correct item: 4. Explanation: Naive Bayes uses conditional independence, making training simpler and more efficient.