Abstract: In realistic decision-making,cost-sensitive issue is one of the important factors which affects human decision-making,and many researchers are committed to reducing the cost of decision-making.At present,in the field of rough set,many researchers mainly research decision-making based on DTRS model and only consider a certain cost,which is not comprehensive enough.While sequential three-way decision model is sensitive to two kinds of costs,and the multi-level granular structure can effectively reduce the total cost of decision and can better simulate the process of human’s dynamic and gradual decision-making.Based on sequential three decision models,this paper constructed a multi-level granular structure.It relates the test cost of each attribute to its classification ability and sets the test cost from the perspective of information entropy.At the same time,combined with the sequential three decisions,the attribute reduction based on the minimum cost criterion is used to remove the influence of redundant attributes and irrelevant attributes on the cost.The experimental results on the seven UCI datasets show that while high accuracy is ensured ,the total cost of decision-making is dropped by an average of 26%,which fully validates the effectiveness of the proposed method.

Key words: Attribute reduction, Cost-sensitive, Multi-granularity, Sequential three-way decision