Density-based Clustering

Density-based clustering aims at finding connected regions of closely spaced points in a data set. The basic idea is to mark core points (points with many surrounding points) and look for clusters in the set of core points. A distance measure is only used to determine a neighborhood for each point. Thus, the choice of a concrete distance measure is not as important as for centroid-based methods or hierarchical clustering.

There exist many ways to fill in the details of the approach. The most widely used density-based clustering algorithms are DBSCAN and OPTICS. Both only assign cluster labels to the training data but do not imply a canonical prediction routine. Both algorithms may yield clusters of arbitrary shape.

Another density-based clustering method, not discussed here, is mean shift.

• DBSCAN
• OPTICS

Related projects:

• Chinese Celadons

  • Hierarchical Clustering

  • Density-based Clustering