Supervised Classification Using Feature Space Partitioning

In the paper we consider the supervised classification problem using feature space partitioning. We first apply heuristic algorithm for partitioning a graph into a minimal number of cliques and subsequently the cliques are merged by means of the nearest neighbor rule. The main advantage of the new approach which optimally utilizes the geometrical structure of the training set is decomposition of the l-class problem (\(l>2\)) into l single-class optimization problems. We discuss computational complexity of the proposed method and the resulting classification rules. The experiments in which we compared the box algorithm and SVM show that in most cases the box algorithm performs better than SVM.