Hamish Burke | 2025-04-03

Related to: #bigData

Clustering Metrics

Dunn index

• Higher it is, better the clustering performance

Silhouette

Measures how well an instance is matched to its cluster

• $a(i)$ is the avg distance between instance $i$ and all other instances in its cluster
• $b(i)$ is the min avg distance between instance $i$ and all other instance in its cluster
• 1 = instance is perfectly clustered
• -1 indicates it should be in a neighbouring cluster
• 0 indicates its on the border of two clusters