Skip to content
uencoder

K-Medoids Clustering

K-Medoids is like K-Means but takes actual data points as centers (medoids), supporting clustering with arbitrary distance metrics.

Purpose

To provide clustering that is more robust to noise and outliers and supports arbitrary distances.

Mechanism

The center (medoid) is the actual data point in the cluster that minimizes total distance. The PAM algorithm swaps a medoid with a non-medoid whenever doing so reduces total distance. It is more robust than K-Means but costlier at O(n^2).

Use cases

  • Robust clustering
  • Non-Euclidean distance clustering
  • Bioinformatics

Related terms

Hierarchical Clustering
Hierarchical clustering builds a nested tree of clusters (dendrogram); the agglomerative approach merges bottom-up while the divisive approach splits top-down.
GLOSSARY

References