Prototype Selection with Compact Sets and Extended Rough Sets

In this paper, we propose a generalization of classical Rough Sets, the Nearest Neighborhood Rough Sets, by modifying the indiscernible relation without using any similarity threshold. We also combine these Rough Sets with Compact Sets, to obtain a prototype selection algorithm for Nearest Prototype Classification of mixed and incomplete data as well as arbitrarily dissimilarity functions. We introduce a set of rules to a priori predict the performance of the proposed prototype selection algorithm. Numerical experiments over repository databases show the high quality performance of the method proposed in this paper according to classifier accuracy and object reduction.