Unsupervised learning algorithms quiz

Question 1

What fundamentally distinguishes unsupervised learning from supervised learning?

Accepted Answer

It works with unlabeled data to find hidden structure

Answer

It uses labeled data to predict outputs

Answer

It always requires a human teacher for every example

Answer

It can only be used for image data

Question 2

What is the main goal of a clustering algorithm?

Accepted Answer

To group similar data points without labels

Answer

To predict a continuous numeric value

Answer

To label each point with a known category

Answer

To reduce the number of training examples

Question 3

In K-means clustering, what does the parameter K represent?

Accepted Answer

The number of clusters to form

Answer

The number of features in the dataset

Answer

The number of iterations to run

Answer

The learning rate of the algorithm

Question 4

Describe the two alternating steps that K-means repeats until convergence.

Accepted Answer

Assign points to nearest centroid, then recompute centroids

Answer

Split and merge clusters

Answer

Sort points then remove outliers

Answer

Train a classifier then predict labels

Question 5

What quantity does K-means try to minimize?

Accepted Answer

The within-cluster sum of squared distances (inertia)

Answer

The number of clusters

Answer

The between-cluster distance only

Answer

The total number of features

Question 6

What is the 'elbow method' used for in clustering?

Accepted Answer

Choosing the number of clusters K

Answer

Choosing the learning rate

Answer

Removing outliers automatically

Answer

Scaling the features

Question 7

Why is feature scaling important before running K-means?

Accepted Answer

Because distance-based methods are biased toward large-range features

Answer

Because it speeds up label prediction

Answer

Because it removes all outliers

Answer

Because K-means cannot handle decimals

Question 8

What problem does K-means++ initialization solve?

Accepted Answer

It selects well-spread initial centroids for better convergence

Answer

It chooses the number of clusters automatically

Answer

It removes the need for scaling

Answer

It guarantees the global optimum every time

Question 9

Why can K-means give different results on different runs with the same data?

Accepted Answer

Because of sensitivity to random initial centroids (local minima)

Answer

Because the data changes each run

Answer

Because it is a supervised method

Answer

Because inertia is random

Question 10

What is a centroid in K-means clustering?

Accepted Answer

The mean position of points in a cluster

Answer

The farthest point in a cluster

Answer

A randomly chosen data point that never moves

Answer

The boundary between two clusters

Question 11

What is hierarchical clustering and what visual output is associated with it?

Accepted Answer

A dendrogram (tree diagram)

Answer

A scatter plot of centroids

Answer

A confusion matrix

Answer

A ROC curve

Question 12

In agglomerative hierarchical clustering, what does a 'linkage' method define?

Accepted Answer

How distance between two clusters is measured

Answer

The number of clusters to keep

Answer

The learning rate for merging

Answer

The random seed used

Question 13

What key advantage does DBSCAN have over K-means?

Accepted Answer

It finds arbitrary-shaped clusters and detects noise without specifying K

Answer

It is always faster than K-means

Answer

It requires labeled data

Answer

It guarantees spherical clusters

Question 14

In DBSCAN, what do the parameters eps and min_samples control?

Accepted Answer

The neighborhood radius and minimum points for a dense region

Answer

The number of clusters and iterations

Answer

The learning rate and momentum

Answer

The number of features and labels

Question 15

How does DBSCAN handle outliers compared to K-means?

Accepted Answer

DBSCAN labels them as noise while K-means forces them into a cluster

Answer

Both ignore outliers equally

Answer

K-means removes them automatically

Answer

Neither can handle outliers

Question 16

What is the primary purpose of Principal Component Analysis (PCA)?

Accepted Answer

To reduce dimensions while retaining maximum variance

Answer

To classify data into labels

Answer

To increase the number of features

Answer

To cluster data into K groups

Question 17

In PCA, what does the first principal component represent?

Accepted Answer

The direction of largest variance in the data

Answer

The direction of smallest variance

Answer

The average of all features

Answer

The cluster centroid

Question 18

Why must principal components be orthogonal (perpendicular) to each other?

Accepted Answer

So components are uncorrelated and capture distinct variance

Answer

To make computation slower

Answer

To force them to be equal in length

Answer

Because labels require it

Question 19

What does 'explained variance ratio' tell you in PCA?

Accepted Answer

The fraction of total variance each component captures

Answer

The error rate of a classifier

Answer

The number of clusters

Answer

The learning rate

Question 20

Which mathematical concepts underlie how PCA finds its components?

Accepted Answer

Eigenvectors and eigenvalues of the covariance matrix

Answer

Derivatives of the loss function

Answer

Gradient descent over labels

Answer

Random forests

Question 21

What is the 'curse of dimensionality' in the context of unsupervised learning?

Accepted Answer

Distances become less meaningful as dimensions grow

Answer

Too few features to train

Answer

Labels become noisy

Answer

The model trains too quickly

Question 22

What is the silhouette score and what range can it take?

Accepted Answer

-1 to +1

Answer

0 to 100

Answer

0 to 1 only

Answer

negative infinity to positive infinity

Question 23

What is the difference between hard clustering and soft (fuzzy) clustering?

Accepted Answer

Hard assigns each point to one cluster; soft gives membership probabilities

Answer

Hard uses labels; soft does not

Answer

Hard is for images; soft is for text

Answer

There is no real difference

Question 24

What does a Gaussian Mixture Model (GMM) assume about the data?

Accepted Answer

Data is a mixture of several Gaussian distributions

Answer

Data comes from one uniform distribution

Answer

Data has no structure

Answer

Data is always one-dimensional

Question 25

What is the goal of association rule mining (e.g., market basket analysis)?

Accepted Answer

Find items that frequently occur together

Answer

Predict a numeric price

Answer

Classify customers into ages

Answer

Reduce data dimensions

Question 26

In association rules, what does 'support' measure?

Accepted Answer

The fraction of transactions containing the itemset

Answer

The accuracy of a classifier

Answer

The number of clusters

Answer

The variance explained

Question 27

In the association rule 'if A then B', what does 'confidence' measure?

Accepted Answer

The probability of B given A

Answer

The total frequency of A and B together

Answer

The number of items in a basket

Answer

The variance of the data

Question 28

What does a 'lift' value greater than 1 indicate in association rule mining?

Accepted Answer

A and B occur together more than expected (positive association)

Answer

A and B are independent

Answer

The rule is invalid

Answer

B never occurs with A

Question 29

What is anomaly (outlier) detection and why is it often unsupervised?

Accepted Answer

Finding points that deviate from normal patterns

Answer

Predicting prices from labels

Answer

Grouping similar customers only

Answer

Increasing data dimensions

Question 30

What is the main purpose of t-SNE?

Accepted Answer

Visualizing high-dimensional data in 2D/3D

Answer

Labeling data automatically

Answer

Speeding up gradient descent

Answer

Generating new data samples

Test yourself