Back to Subject

Unit 1: Introduction to Unsupervised Learning & K-Means - Practice Quiz

INT423 — Machine Learning-Ii 50 Questions
0 Correct 0 Wrong 50 Left
0/50

1 What is the primary characteristic of Unsupervised Learning?

A. The algorithm trains on labeled data
B. The algorithm predicts a continuous numerical value
C. The algorithm uses a feedback loop for rewards
D. The algorithm trains on data without labels

2 Which of the following is a primary goal of clustering?

A. To predict future values based on past trends
B. To reduce the noise in a signal
C. To classify images into predefined categories
D. To group similar data points together

3 In the context of K-Means, what does 'K' represent?

A. The number of data points
B. The number of iterations
C. The dimension of the features
D. The number of clusters

4 What kind of problem is K-Means designed to solve?

A. Reinforcement Learning
B. Classification
C. Clustering
D. Regression

5 What is a 'centroid' in the K-Means algorithm?

A. The geometric center of a cluster
B. The data point furthest from the center
C. An outlier in the dataset
D. The boundary line between clusters

6 Which distance metric is most commonly used in standard K-Means?

A. Manhattan distance
B. Euclidean distance
C. Cosine similarity
D. Hamming distance

7 What is the first step of the K-Means algorithm?

A. Calculate the total error
B. Assign points to the nearest cluster
C. Update the centroids
D. Initialize cluster centroids

8 During the assignment step of K-Means, how is a data point assigned to a cluster?

A. To the cluster with the most points
B. To the cluster with the closest centroid
C. Randomly
D. To the cluster with the highest variance

9 What happens during the update step of the K-Means algorithm?

A. New data points are added
B. Centroids are moved to the mean of their assigned points
C. The number of clusters (K) is increased
D. Points are reassigned to different clusters

10 When does the K-Means algorithm stop iterating?

A. After exactly 10 iterations
B. When K is equal to N
C. When the centroids do not change significantly
D. When the training error is zero

11 What is the optimization objective (cost function) of K-Means?

A. Maximize the Silhouette score
B. Minimize Within-Cluster Sum of Squares (WCSS)
C. Minimize the number of clusters
D. Maximize Inter-cluster distance

12 The objective function of K-Means is non-convex. What does this imply?

A. It always finds the global minimum
B. It cannot be optimized
C. It requires labeled data
D. It may get stuck in a local minimum

13 Which of the following is a disadvantage of the K-Means algorithm?

A. It works only on labeled data
B. It cannot handle numerical data
C. It is computationally very expensive for small datasets
D. It is sensitive to outliers

14 If you set K equal to the number of data points (N), what will the WCSS be?

A. Zero
B. Maximum possible value
C. Infinity
D. Undefined

15 What is the 'Elbow Method' used for?

A. Speeding up convergence
B. Initializing centroids
C. Handling outliers
D. Determining the optimal number of clusters (K)

16 In the Elbow Method plot, what is typically on the Y-axis?

A. Number of clusters (K)
B. Inertia or WCSS
C. Time taken
D. Accuracy

17 What is the 'Random Initialization Trap' in K-Means?

A. The algorithm fails if data is random
B. Random data points cannot be clustered
C. Randomly picking centroids can lead to poor local optima
D. Choosing K randomly leads to errors

18 What is K-Means++?

A. A post-processing step for K-Means
B. A smarter initialization technique for K-Means
C. A version of K-Means for supervised learning
D. A method to choose the optimal K

19 How does K-Means++ select the first centroid?

A. It picks one data point uniformly at random
B. It calculates the global mean
C. It chooses the point furthest from the origin
D. It picks the point with the highest variance

20 What is the difference between Hard Clustering and Soft Clustering?

A. Hard clustering assigns a point to one cluster; Soft assigns probabilities
B. Hard clustering is faster; Soft is slower
C. Hard clustering allows overlapping; Soft does not
D. Hard clustering uses K-Means; Soft uses Decision Trees

21 Standard K-Means is an example of which type of clustering?

A. Hard Clustering
B. Soft Clustering
C. Hierarchical Clustering
D. Density-based Clustering

22 Which algorithm is a well-known example of Soft Clustering?

A. Agglomerative Clustering
B. DBSCAN
C. K-Means
D. Fuzzy C-Means

23 If a data point has a membership vector [0.7, 0.2, 0.1] for 3 clusters, this is an example of:

A. Soft Clustering
B. Hard Clustering
C. Regression
D. Outlier Detection

24 What shape of clusters does K-Means typically assume?

A. Arbitrary shapes
B. Spherical or convex
C. Elongated shapes
D. Spirals

25 Why is feature scaling (standardization/normalization) important in K-Means?

A. It is not important
B. To ensure the algorithm runs faster only
C. To convert categorical data to numerical
D. To prevent features with larger ranges from dominating the distance metric

26 What is the computational complexity of one iteration of K-Means?

A. O(e^N)
B. O(N^2)
C. O(N * log N)
D. O(K N d)

27 In the Elbow method, the 'elbow' point represents:

A. The point where K equals 1
B. The point where adding another cluster does not significantly reduce WCSS
C. The point where WCSS becomes zero
D. The point of maximum error

28 Which of the following implies that K-Means has converged?

A. The number of clusters decreases
B. The data becomes labeled
C. The assignment of points to clusters remains unchanged
D. WCSS increases

29 What is 'Inertia' in the context of Scikit-Learn's K-Means implementation?

A. The distance between cluster centers
B. The number of iterations
C. The time taken to run
D. The sum of squared distances of samples to their closest cluster center

30 Which strategy is used to mitigate the local optima problem in K-Means?

A. Decrease the learning rate
B. Run the algorithm multiple times with different initializations
C. Increase the number of clusters
D. Use Manhattan distance

31 Can K-Means handle categorical data directly?

A. Yes, using Hamming distance
B. Yes, it works natively
C. Only if the data is ordinal
D. No, it requires numerical data

32 In K-Means++, how is the probability of selecting the next centroid determined?

A. Randomly with uniform distribution
B. Proportional to the squared distance from the nearest existing centroid
C. Inversely proportional to the distance from existing centroids
D. Based on the density of the points

33 What is a 'Voronoi Diagram' in relation to K-Means?

A. A method to initialize K
B. A type of soft clustering
C. A visualization where regions are defined by the closest centroid
D. A plot of the cost function

34 If the clusters in the data are of very different densities and sizes, K-Means will:

A. Perform perfectly
B. Merge the clusters
C. Automatically adjust the metric
D. Likely fail to identify the correct clusters

35 Which step ensures K-Means is an unsupervised algorithm?

A. Calculating the mean
B. Iterating until convergence
C. Not using target labels for training
D. Minimizing WCSS

36 In the equation for WCSS, what is being squared?

A. The distance between a point and its assigned centroid
B. The distance between two centroids
C. The number of clusters
D. The number of iterations

37 Why is it often difficult to pick the optimal K using the Elbow method?

A. It requires labeled data
B. The 'elbow' might not be sharp or clear
C. The plot is always a straight line
D. It takes too long to compute

38 What is the primary role of the 'Coordinate Descent' concept in K-Means?

A. It is used for initialization
B. It is the method used to optimize the objective function
C. It calculates the distance
D. It is used to visualize data

39 If you perform K-Means on a dataset with 2 distinct well-separated blobs but set K=4, what happens?

A. It finds 2 clusters and ignores the other 2
B. It merges the blobs
C. It splits the natural blobs into smaller clusters
D. The algorithm crashes

40 In Soft Clustering, the sum of membership weights for a single data point across all clusters usually equals:

A. 100
B. 0
C. K
D. 1

41 Which of the following is NOT an application of K-Means?

A. Customer Segmentation
B. Spam Classification (Supervised)
C. Document Clustering
D. Image Compression (Color Quantization)

42 Does K-Means guarantee finding the global optimum for the WCSS?

A. No, it depends on initialization
B. Yes, always
C. Yes, if K is small
D. Only if using Manhattan distance

43 The computational cost of the distance calculation step for one point against K centroids is proportional to:

A. N
B. K
C. 1
D. N^2

44 Which component constitutes the 'model' after training K-Means?

A. The coordinates of the final centroids
B. The original dataset
C. The Elbow plot
D. The list of outliers

45 What is the relationship between Within-Cluster variance and Between-Cluster variance in a good clustering?

A. High within-cluster, Low between-cluster
B. Low within-cluster, Low between-cluster
C. Low within-cluster, High between-cluster
D. High within-cluster, High between-cluster

46 Lloyd's Algorithm is another name for:

A. K-Means Algorithm
B. Hierarchical Clustering
C. DBSCAN
D. KNN

47 In the context of image segmentation, what does a pixel represent in K-Means?

A. A label
B. A data point
C. A centroid
D. A cluster

48 Why might one choose a K value slightly different from the Elbow point?

A. Because the Elbow method is always wrong
B. To increase computational cost
C. Based on business requirements or downstream tasks
D. To maximize WCSS

49 If K=1, the centroid location will be:

A. Undefined
B. A random data point
C. The origin (0,0)
D. The mean of the entire dataset

50 What happens if a cluster becomes empty during K-Means iterations?

A. The K value increases
B. It is ignored and WCSS becomes 0
C. The algorithm stops
D. The empty cluster is usually re-initialized or removed
Next MCQ
Unit 2