DSE Unsupervised Learning: Clustering 2 — Questions and Answers
Question 1: Which internal cluster validation metric measures the ratio of intra-cluster distances to inter-cluster distances?
- Davies-Bouldin Index (Correct answer)
- Silhouette Score
- Dunn Index
- Calinski-Harabasz Index
Correct answer: Davies-Bouldin Index
The Davies-Bouldin Index measures the average similarity ratio of each cluster to its most similar cluster, where lower values indicate better clustering.
Question 2: In DBSCAN, a point that is reachable from a core point but is not itself a core point is called a:
- Noise point
- Border point (Correct answer)
- Centroid point
- Leaf point
Correct answer: Border point
Border points lie within the epsilon neighborhood of a core point but do not have enough neighbors to be core points themselves.
Question 3: What is the primary advantage of using Mini-Batch K-Means over standard K-Means?
- Produces better cluster quality
- Handles non-spherical clusters
- Faster convergence on large datasets (Correct answer)
- Automatically selects k
Correct answer: Faster convergence on large datasets
Mini-Batch K-Means uses small random subsets of data in each iteration, dramatically reducing computation time while maintaining comparable quality.
Question 4: Which linkage criterion in hierarchical clustering tends to produce compact, spherical clusters?
- Single linkage
- Complete linkage
- Average linkage
- Ward linkage (Correct answer)
Correct answer: Ward linkage
Ward linkage minimizes the total within-cluster variance at each merge step, consistently producing compact and similarly-sized clusters.
Question 5: A data scientist runs K-Means with k=5 and observes that one cluster has only 2 points out of 10,000. What is the most likely issue?
- The algorithm converged too early
- Poor initialization of centroids (Correct answer)
- Too many iterations
- The silhouette score is too high
Correct answer: Poor initialization of centroids
Poor centroid initialization can trap a centroid in a sparse region, resulting in a near-empty cluster while others are overcrowded.
Question 6: In spectral clustering, what is the role of the Laplacian matrix?
- Computes pairwise Euclidean distances
- Encodes graph connectivity for eigendecomposition (Correct answer)
- Normalizes feature values
- Selects the number of clusters automatically
Correct answer: Encodes graph connectivity for eigendecomposition
The graph Laplacian captures the connectivity structure of the data, and its eigenvectors reveal cluster membership in the embedded space.
Question 7: Which of the following clustering algorithms is best suited for discovering clusters of arbitrary shape in spatial data?
- K-Means
- Gaussian Mixture Models
- DBSCAN (Correct answer)
- Agglomerative clustering with Ward linkage
Correct answer: DBSCAN
DBSCAN defines clusters by density rather than distance to centroids, enabling it to find clusters of any shape including rings and crescents.
