Feature ranking in rough sets

摘要

The paper proposes a novel feature ranking technique using discernibility matrix. Discernibility matrix is used in rough set theory for reduct computation. By making use of attribute frequency information in discernibility matrix, the paper develops a fast feature ranking mechanism. Based on the mechanism, two heuristic reduct computation algorithms are proposed. One is for optimal reduct and the other for approximate reduct. Empirical results are also reported.