K-MEANS CLUSTERING
Algorithm Overview:

• Initialization: Choose K centroids
  
  
• Recursion:
  
  
  • Assign all vectors to their nearest neighbor.
    
    
  • Recompute the centroids as the average of all vectors assigned to the same centroid.
  
  
• Termination: Check overall distortion.

For a typical implementation of K-MEANS, see our pattern recognition applet.

Issues:

• Distance measure: Euclidean? Mahalanobis?
  
  
• Centroid computation: Average? Median? Min-Max?
  
  
• Splitting/Merging: Sparsity? Separability?
  
  
• Number of clusters: When do we stop?