A VIEW INSIDE THE CLASSIFICATION WITH NON-NESTED GENERALIZED EXEMPLARS

This paper analyses the impact on the classification accuracy of three elements of the Non-Nested Generalized Exemplars (NNGE) classifier: the hyperrectangles splitting procedure, the pruning of non-generalized exemplars and the presentation order of training instances. As a consequence of this analysis some NNGE variants are proposed. A statistical analysis reveals that the proposed variants improve the classification ability of the original NNGE.