A Fast Reduction Algorithm with Attribute Pre-Sort Based on Neighborhood Rough Set

Abstract

Neighborhood rough set (NRS) is a classical extension model of Pawlak rough set, which has been used to evaluate the importance of attributes for attribute reduction. In addition to attribute evaluation, attribute search strategy is also a very important issue for attribute reduction. In this paper, we define the concentration, dispersion, stability degree of samples with respect to the single attribute to measure the significance of attributes, and use the stability of samples to pre-sort attributes. A fast attribute reduction algorithm with attribute pre-sort based on neighborhood rough set (APNRS) is designed to search a reduct, and the reduct is more conducive to classify learning tasks. Compared with the traditional greedy search algorithms, the APNRS algorithm greatly improves the computational efficiency under the condition of ensuring classification accuracy. Finally, a series of numerical experiments are carried out to verily the efficiency of the APNRS.