Abstract: The problems existed in the Pawlak attribute importance based method by which the attribute weight is constructed were analyzed in detail firstly,then a discernibility matrix based definition of attribute importance was given.On this basis,a novel approach for constructing the combinatorial attribute weights of information systems was proposed.In this approach,the attribute weights are ascertained according to the condition attribute’s contribution in the whole information system.The proposed approach not only reflects the ability of condition attributes to distinguish the object,but also reflects the classification capability of each attribute in the whole condition attributes.The numerical example demonstrates that the attribute weights gained by the proposed approach are more closer to the facts,so the proposed approach can improve the accuracy of the attribute weights.

Key words: Rough sets,Dscernibility matrix,Attribute significance,Weight