Data Reduction via Conflicting Data Analysis

Abstract

This paper introduces a new method for instances selection. The conceptual framework and the basic notions used by this method are those of an extended rough set theory, called α-rough set theory. In this context we formalize a notion of conflicting data, which is at the basis of a conflict normalization method used for instances selection. Extensive experiments are performed to show the efficiency and the accuracy of models built from the reduced datasets. The selection methodology and its results are discussed.