15.4: Current Research Directions

The most significant problems associated with clustering now are associated with scaling existing algorithms cleanly with two attributes: size and dimensionality. To deal with larger and larger datasets, algorithms such as canopy clustering have been developed, in which datasets are coarsely clustered in a manner intended to pre-process the data, following which standard clustering algorithms (e.g. k-means) are applied to sub- divide the various clusters. Increase in dimensionality is a much more frustrating problem, and attempt to remedy this usually involve a two stage process in which appropriate relevant subspaces are first identified by appropriate transformations on the original space and then subjected to standard clustering algorithms.

---