Next: Clustering by graph theory
Up: Partitional Clustering
Previous: Clustering by mixture decomposition
- Clusters are viewed as regions of the pattern space in which the
patterns are dense, separated by regions of low pattern density.
- Clusters can be identified by searching for regions of high
density, called modes, in the pattern space.
- Each mode is associated with the cluster center.
- Each pattern is assigned to the cluster with the closest center.
- Probability density estimate at a point x is proportional to the number of
patterns falling in a small region around x.
The choice of is critical when n is small.
- Simplest way to identify modes is to construct a histogram by
partitioning the pattern space into a number of non-overlapping
regions. Number of patterns should be large to get a good
estimate.
Miranda Maria Irene
Thu Apr 1 15:43:18 IST 1999