Overview of clustering algorithms

A survey of clustering algorithms is presented, paying particular attention to those algorithms that require the least amount of a priori knowledge about the domain being clustered.

Abstract

Clustering algorithms are useful whenever one needs to classify an excessive amount of information into a set of manageable and meaningful subsets. Using an analogy from vector analysis, a clustering algorithm can be said to divide up state space into discrete chunks such that each vector lies within one chunk. These vectors can best be thought of as sets of features. A canonical vector for each region of state space is chosen to represent all vectors which are located within that region. The following paper presents a survey of clustering algorithms. It pays particular attention to those algorithms that require the least amount of a priori knowledge about the domain being clustered. In the current work, an algorithm is compelling to the extent that it minimizes any assumptions about the distribution of vectors being classified.