Rough K-medoids clustering using GAs

Abstract

This paper proposes a medoid based variation of rough K-means algorithm. The variation can be especially useful for a more efficient evolutionary implementation of rough clustering. Experimentation with the rough K-means algorithm has shown that it provides a reasonable set of lower and upper bounds for a given dataset. However, rough K-means algorithm has not been explicitly shown to provide optimal rough clustering. Recently, an evolutionary rough K-means algorithm was proposed that minimizes a rough within-group-error. The proposal combined the efficiency of rough K-means algorithm with the optimization ability of GAs. The medoid based variation proposed here is more efficient than the evolutionary rough K-means algorithm, as it uses a smaller and discrete search space. It will also make it possible to test a wider variety of optimization criteria due to built in restrictions on the solution space.