Multi-Agent Evolutionary Clustering Algorithm Based on Manifold Distance

Abstract

By using the manifold distance as the similarity measurement, a multi-agent evolutionary clustering algorithm based on manifold distance (MAEC-MD) is proposed in this paper. MAEC-MD designs a new connection based encoding, and the clustering results can be obtained by the process of decoding directly. It does not require the number of clusters to be known beforehand and overcomes the dependence of the domain knowledge. Aim at solving the clustering problem, three effective evolutionary operators are designed for competition, cooperation, and self-learning of an agent. Some experiments about artificial data, UCI data are tested. These results show that MAEC-MD can confirm the number of clusters automatically, tackle the data with different structures, and satisfy the diverse clustering request.