Computer scienceData scienceMachine learningClusteringClustering evaluation

Introduction to clustering evaluation

Opting out of silhouette

Report a typo

Suppose there is a dataset which looks like this:

A scatter plot of data with noise and non-spherical potential clusters

Explain why using the silhouette score might yield low scores regardless of the algorithm used.

Select one or more options from the list

By the visual observation, the data is noisy, and the silhouette score is sensitive to noise

The clusters don't appear to be universally spherical, which is the underlying metric assumption

The densities appear to be similar, and the silhouette score gives a higher score to the clusters with various densities

The clusters aren't well-separated, which is the underlying metric assumption

___

Create a free account to access the full topic