Evolutionary Clustering Algorithms for Relational Data

Abstract

Relational data clustering has received lot less attention than vector data clustering and the use of evolutionary techniques to optimize clustering parameters is even rare. We extend an earlier work where a relational data version of DBSCAN was presented and an evolutionary framework was proposed for optimizing clustering parameters. Five evolutionary techniques are presented in this paper – three algorithms based on particle swarm optimization, the firefly algorithm and the composite differential evolution technique. Clustering results from the proposed methodologies are tested on benchmark datasets from the UCI machine learning database.