An ant-based new clustering model for graph proximity construction

Abstract

This paper presents a new concept for an artificial ant model to build proximity graphs. We tried first to introduce the state of art of different clustering methods relying on the swarm intelligence and the ants numerous abilities. Our new bio-inspired model is based on artificial ants over a dynamic graph of clusters using colonial odours and pheromone-based reinforcement process. Our non-hierarchical algorithm, called CL-Ant, where each ant represents one datum and its moves aim to create homogeneous data groups that evolve together in a proximity graph environment. In this model, the artificial ant performs two steps: following the maximum pheromone path rate, and then, integrating to neighbours clusters using simple localisation rules. Afterwards we present an incremental extension, called CL-AntInc to treat data streams, which allows building graphs in an incremental way. Our survey properties were studied thoroughly and a detailed analytical comparison of our results with those obtained by other methods was provided. These algorithms were evaluated and validated using real databases extracted from the Machine Learning Repository.