A Rough Set Algorithm for Attribute Reduction via Mutual Information and Conditional Entropy

Attribute reduction is one of the kernel components in the rough set theory, which has been successfully applied in many fields. This paper firstly analyzes the major drawbacks that most of the current algorithmic approaches hold. Then it proposes a novel heuristic algorithm, which redefines the significance function of attributes using mutual information and conditional entropy. Also this paper considers the redundancy variation of the existing attributes within the reduct set influenced by the newly-added elements. Besides, the attribute dependency measurement is modified instead of common entropy-based assessment during the procedure in this article. The experimental result shows that our algorithm can obtain the reduct set with minimal number of members under most circumstances with faster convergence rate than other entropy-based methods.