Evolutionary pruning of non-nested generalized exemplars

This paper investigates the ability of an evolutionary pruning mechanism to improve the predictive accuracy of a classifier based on non-nested generalized exemplars. Two pruning algorithms are proposed: one which selects the most representative generalized exemplars and the other one which simultaneously selects both relevant exemplars and relevant attributes. Experimental studies conducted for a set of twenty-one datasets illustrated that both algorithms induce a significant improvement on the classification ability of the selected set of non-nested generalized exemplars.