Distance Based On Neighborhood Classifier and Attribute Reduction

Abstract

In the neighborhood rough set model, with the increasing of the size of information granules, the neighborhood classifier based on the majority voting rule is easy to misjudge the classes of testing samples. To remedy this deficiency, a strategy of attribute reduction based on the idea of the minimum mean distance is proposed in this paper. Firstly, a neighborhood relation is presented by a distance function. Secondly, neighborhood mean distance classifier judges the class of testing sample with the minimum mean distance instead of the majority voting rule in the decision system. Finally, the experimental results on 8 UCI data sets tell us that the reduct obtained by our strategy can not only decrease the conditional entropy, but also provide better classification performances in larger scale information granules.