Ant Colony Optimisation-Based Classification Using Two-Dimensional Polygons

The application of Ant Colony Optimization to the field of classification has mostly been limited to hybrid approaches which attempt at boosting the performance of existing classifiers (such as Decision Trees and Support Vector Machines (SVM)) often through guided feature reductions or parameter optimizations. In this paper we introduce Po1yACO: A novel Ant Colony based classifier operating in two dimensional space that utilizes ray casting. To the best of our knowledge, our work is the first reported Ant Colony based classifier which is non-hybrid, in the sense, that it does not build on any legacy classifiers. The essence of the scheme is to create a separator in the feature space by imposing ant-guided random walks in a grid system. The walks are self-enclosing so that the ants return back to the starting node forming a closed classification path yielding a many edged polygon. Experimental results on both synthetic and real-life data show that our scheme is able to perfectly separate both simple and complex patterns, without utilizing "kernel tricks" and outperforming existing classifiers, such as polynomial and linear SVM. The results are impressive given the simplicity of Po1yACO compared to other approaches such as SVM.