A Weighting Approach for Features Based on Real Rough Set

For the computation of feature weights, this paper proposed a novel approach through introducing the concept of feature significance degree on the real rough set theory. The feature significance degree could characterize the contribution of features to the decision making more objectively. The proposed approach was applied to the benchmark test sets Reuters-21578 Top10 and 20 Newsgroups to examine its effectiveness. The results show that the proposed approach improves the distribution status of sample space. It makes the samples in the same class more compact and those in different classes looser.