Abstract: The attribute reduction is one of the key issues on knowledge acquisition in rough set theory．In this paper,the discernibility matrix was utilized to obtain the core attribute first,then a new heuristic algorithm for attribute reduction was proposed based on the importance ratings of the attribute．Our algorithm can give proper combination of attri-butes for effective attribute reduction and therefore the complex computing is avoided,which is different from some methods relying on algebra and information entropy．Finally,the example analysis demonstrates the validity and feasibi-lity of our algorithm.

Key words: Data mining,Attribute reduction,Algorithm,Rough set