Machine Learning Clustering 5 — Questions and Answers
Question 1: Which distance metric is most appropriate for clustering text documents represented as TF-IDF vectors?
- Euclidean distance
- Manhattan distance
- Cosine similarity (Correct answer)
- Hamming distance
Correct answer: Cosine similarity
Cosine similarity measures the angle between vectors, making it robust to document length differences common in TF-IDF representations.
Question 2: What is the primary advantage of using mini-batch K-Means over standard K-Means?
- It always finds the global optimum
- It handles categorical data natively
- It scales to very large datasets by using random subsets per iteration (Correct answer)
- It automatically determines the optimal K
Correct answer: It scales to very large datasets by using random subsets per iteration
Mini-batch K-Means updates centroids using random subsets (mini-batches) of data, making it much faster on large datasets with minimal accuracy loss.
Question 3: When the Davies-Bouldin Index is used to evaluate clustering quality, what does a lower value indicate?
- More clusters were used
- Poorer cluster separation
- Better clustering with compact and well-separated clusters (Correct answer)
- Higher inertia
Correct answer: Better clustering with compact and well-separated clusters
A lower Davies-Bouldin Index means clusters are more compact internally and more separated from each other, indicating better clustering.
Question 4: Which statement about the K-Medoids algorithm is correct compared to K-Means?
- K-Medoids is faster than K-Means on large datasets
- K-Medoids centroids must be actual data points (Correct answer)
- K-Medoids minimizes the sum of squared distances
- K-Medoids cannot handle outliers
Correct answer: K-Medoids centroids must be actual data points
In K-Medoids, the representative of each cluster (medoid) must be an actual data point, making it more robust to outliers than K-Means.
Question 5: What does a silhouette score close to -1 for a data point indicate?
- The point is perfectly clustered
- The point is on the boundary between two clusters
- The point is likely assigned to the wrong cluster (Correct answer)
- The point is an outlier detected by DBSCAN
Correct answer: The point is likely assigned to the wrong cluster
A silhouette score near -1 means the point is much closer to a neighboring cluster than its own, suggesting a misclassification.
Question 6: In hierarchical clustering, which linkage method tends to produce the most balanced cluster trees?
- Single linkage
- Complete linkage
- Ward linkage (Correct answer)
- Average linkage
Correct answer: Ward linkage
Ward linkage minimizes the total within-cluster variance at each merge step, typically producing balanced, compact, and similarly-sized clusters.
Question 7: Which clustering approach would be most appropriate for customer segmentation when the true number of segments is unknown and data has irregular shapes?
- K-Means with K=5
- DBSCAN with tuned epsilon and minPts (Correct answer)
- Gaussian Mixture Model with diagonal covariance
- Hierarchical clustering with single linkage
Correct answer: DBSCAN with tuned epsilon and minPts
DBSCAN automatically determines the number of clusters and handles arbitrary shapes, making it suitable when K is unknown and cluster geometry is complex.
Machine Learning Clustering 5
Which distance metric is most appropriate for clustering text documents represented as TF-IDF vectors?