Distance Based Clustering

Which of these distance measures will lead to a clustering algorithm that is more sensitive to outliers?

Why is it important to take the best solution over several replications of K-means (or K-medoids) clustering?

• To make sure all of the data points are used in the calculation.

• To account for numerical errors in the computations.

• To find a global minimum of the non-convex objective function.

• To account for uncertainty in the observations.

• single (nearest-neighbour)

• complete (furthest-neighbour)

• average

Examine this simple dendrogram. How many clusters would be obtained if we cut the tree at height h=4.5?

• 1 cluster

• 2 clusters

• 3 clusters

• 4 clusters

Consider this plot of number of clusters against average silhouette score. What is the optimal number of clusters suggested by this metric?

• 3

• 4

• 5

• 6