On Data Clustering with Artificial Ants

Contents

We present in this paper a new ant based approach named AntClass for data clustering. This algorithm uses the stochastic principles of an ant colony in conjunction with the deterministic principles of the Kmeans algorithm. It first creates an initial partition using an improved ant-based approach, which does not require any information on the input data (such as the number of classes, or an initial partition). Then it uses the Kmeans to speed up the convergence of the stochastic approach. In a second phase, AntClass uses hierarchical clustering where ants may cluster together heaps of objects and not just objects. We also use an heterogeneous population of ants in order to avoid complex parameter settings. We show on typical benchmark databases that AntClass is competitive with other approaches such as the Kmeans or ISODATA.