Clusters as Galaxies in Data Space
- Each map cell can be thought of as a "unit-mass star" in an
N-dimensional space
- The density of "stars" varies throughout space often forming
clumps or "galaxies"
- The clustering task is to determine, in an iterative fashion,
which "stars" belong together in a "galaxy"
- Cluster centroids or "galactic centers-of-mass" are recomputed
at the end of each iteration based on assignment of "stars" to
"galaxies"
- So the "centers-of-mass" migrate toward the most densely
populated regions of "data space"
- The procedure stops when very few "stars" change "galactic"
assignment
- The final cluster centroids represent the average combination of
conditions for the N variables for an individual cluster
- The number of clusters or "galaxies" is specified by the
user