Abstract
A method, called HACO2 (Hyperbox classification with Ant Colony Optimization), is proposed for evolving a classifier for labeled data using hyperboxes and an ant colony meta-heuristic. It reshapes the hyperboxes in a near-optimal way, improving the accuracy of the results while acknowledging the topological information (inherently associated to classification) of the data. It also allows a feature discriminating ratio to determine which characteristics are more important for distinguishing classes. The method is validated using artificial 2D data and then applied to the benchmark iris database. Both experiments provide results with over 95% accuracy. Further modifications (automatic parameter setting) and extensions (initialization short comes) and applications to the field of software assessment are discussed.
