Properties Of Potential Function- Based Clustering Algorithms

Abstract

The clustering algorithms based on potential functions are capable of clustering a set of data, making no implicit assumptions on the cluster shapes and without knowing in advance the number of clusters. They are similarity-based type clustering algorithms and do not use any prototype vectors of the clusters. In this paper, some properties of these algorithms are studied: points arrangement tendency, constant potential surface, cluster shapes and robustness to noise.

Keywords

• potential function
• clustering algorithm
• measure of similarity