A PROPOSITION OF THE NEW FEATURE SPACE AND ITS USE TO CONSTRUCTION OF A FAST MINIMUM DISTANCE CLASSIFIER

The paper presents a new approach to the reduction of the reference set size for minimum distance classifiers. An idea of the proposed method consists in mapping of each point from the reference set into a new feature space. Feature selection in the new feature space is equivalent to reduction of a number of hyperplane pieces which form the separating hypersurface. An effectiveness of the considered approach is shown on a simple artificial example. Furthermore, it is verified on a real classification problem.