A Heuristic Algorithm for Attribute Reduction Based on the Discernibility Matrix
Qibing Zhu · Computer Engineering and Science · 2008
In order to obtain good relative attribute reduction in decision systems, a heuristic algorithm for attribute reduction based on discernibility matrix is proposed. The algorithm is based on the discernibility matrix, not only the mutual information between selected conditional attributes and decision attributes are considered, but also its value distribution. A new attribute importance measurement method is defined from the viewpoint of information theory, and the measurement is used as the heuristic information. Finally an attribute reduction set is obtained. The experimental results show that the algorithm can effectively reduce the decision system and obtain ideal reduction results, and that the number of decision rules after the reduction is small.