MLS-C01 Question 118
Select 3You are tasked with performing a cluster analysis on a dataset containing customer purchase behaviors. Before applying a clustering algorithm, you decide to determine the optimal number of clusters. Which of the following techniques would help you achieve this?
- A
Use the elbow plot to analyze the within-cluster sum of squares (WCSS) across different cluster counts.
- B
Perform hierarchical clustering and analyze the dendrogram to identify natural cluster divisions.
- C
Apply k-means clustering with a fixed number of clusters and monitor the runtime performance.
- D
Use the silhouette score to evaluate the quality of clustering for different numbers of clusters.
- E
Examine the size of clusters after fitting the clustering algorithm to ensure they are evenly distributed.
Show answer and explanation
Correct answers: A, B, D
Explanation
Determining the optimal number of clusters is critical in cluster analysis. The elbow plot helps find the point of diminishing returns in WCSS. Hierarchical clustering's dendrogram provides a visual representation of natural cluster groupings. The silhouette score evaluates cluster quality and can indicate the ideal number of clusters. However, monitoring runtime performance or cluster size distribution does not directly contribute to determining the optimal cluster count.
- A. Correct.
This is correct. The elbow plot is a common method to determine the optimal number of clusters by identifying the 'elbow point' where the WCSS reduction slows down significantly.
- B. Correct.
This is correct. Hierarchical clustering produces a dendrogram, which can help visualize and decide the natural grouping and the number of clusters.
- C. Incorrect.
This is incorrect. While k-means can be applied with a fixed number of clusters, monitoring runtime performance does not help in determining the optimal number of clusters.
- D. Correct.
This is correct. The silhouette score measures how similar an object is to its cluster compared to other clusters, helping evaluate the quality of clustering for different cluster counts.
- E. Incorrect.
This is incorrect. Examining cluster sizes post-clustering is useful for analysis but does not directly help in determining the optimal number of clusters.