Hamish Burke | 2025-04-03

Related to: #bigData

Hierarchical clustering

• don't have to commit to number of clusters beforehand
• creates a tree-based representation called a dendrogram
• Agglomerative/bottom-up
  • Start with leaves of tree
  • Merge most similar clusters until left with one
• Dissimilarity measure
  • Euclidean distance $\sqrt{(a-b{)}^2}$
  • distance squared $(a-b{)}^2$
  • Manhattan distance
  • Correlation measure $1-corr(a,b)$
• Linkage methods
  • Single
    • Minimum distance
  • Complete
    • Maximum distance
  • Average
    • All distances, averaged
  • Centroid
    • Based off mean of cluster