Aspects of Fuzzy Parameter Tuning for Partitive Clustering Algorithms Based on Potential Functions

Abstract

In clustering algorithms, potential functions can be used as a measure of similarity, characterizing the membership of a point to a group of points. Potential function-based algorithms (PFBA) are suited for clustering of complex-shape data sets, without knowing in advance the number of clusters. Also, they make no implicit assumptions on the cluster shapes and do not use any prototype vectors of the clusters. The parameter of potential function has direct influence on clustering performance. In this paper, some expert rules of PFBA are generated for using in selecting process of parameter values. A fuzzy system for parameter tuning of PFBA is studied, so that the best clustering to be obtained with less seeking efforts.