1. Question 1

Telecom predicts service cancellations → classification problem. Which model?

❌ Decision trees
❌ Neural networks
❌ Naïve Bayes
✅ K-nearest neighbors

Explanation:
All models can be used, but KNN is the intended answer here because it is commonly introduced as a simple, effective classification model based on past behavior similarity.

2. Question 2

One-vs-One classification strategy uses which method for final class?

❌ Maximal margin vote
✅ Popularity vote
❌ Confidence-based ranking
❌ Probability average

Explanation:
One-vs-one trains multiple binary classifiers. Each classifier “votes” for a class, and the class with the most votes wins.

3. Question 3

In a decision tree, entropy measures:

✅ The level of disorder or randomness in a node
❌ Count of final nodes
❌ Average feature value
❌ Depth of the tree

Explanation:
Entropy quantifies impurity — how mixed the classes are at a node.

4. Question 4

Regression tree on continuous features — which splitting method does not scale well?

❌ Midpoints method
❌ MSE method
❌ Entropy reduction method
✅ Exhaustive search method

Explanation:
Exhaustive search evaluates every possible split, which becomes extremely slow on large datasets.

5. Question 5

Why does KNN accuracy drop when K increases?

❌ Scaling errors
✅ Too much smoothing of patterns
❌ Small training data
❌ Many irrelevant features

Explanation:
Large K averages over too many neighbors, blurring class boundaries → underfitting.

6. Question 6

Adjusting epsilon (ɛ) in SVR controls:

❌ Number of support vectors
✅ Maximum allowed error within the margin (epsilon tube)
❌ Kernel choice
❌ Decision boundary complexity

Explanation:
Epsilon defines the width of the tube where errors are ignored.

7. Question 7

What is the primary goal of AdaBoost?

❌ Reduce overfitting with deep trees
✅ Create a strong learner from weak learners by reducing bias
❌ Combine models in parallel
❌ Dimensionality reduction

Explanation:
AdaBoost trains weak learners sequentially, each focusing on correcting prior errors.

🧾 Summary Table

Q#	Correct Answer	Key Concept
1	K-nearest neighbors	Classification using similarity
2	Popularity vote	One-vs-One strategy
3	Disorder/randomness	Entropy in decision trees
4	Exhaustive search	Splitting continuous features
5	Too much smoothing	High K → underfitting
6	Max allowed error (epsilon tube)	SVR epsilon parameter
7	Strong learner from weak learners	AdaBoost goal

Graded Quiz: Building Supervised Learning Models :Machine Learning with Python (IBM AI Engineering Professional Certificate) Answers 2025