Attribute Reduction Based on Multicost Decision-theoretic Rough Set

DOI：10.11896/j.issn.1002-137X.2017.09.013

 作者 单位 E-mail 杨志荣 江苏科技大学计算机科学与工程学院 镇江212003 qimuyunduan@163.com 王宇 江苏科技大学计算机科学与工程学院 镇江212003 liuyuedewangchao@163.com 杨习贝 江苏科技大学计算机科学与工程学院 镇江212003南京理工大学经济管理学院 南京210094 zhenjiangyangxibei@163.com

与经典粗糙集相比,传统的决策粗糙集将代价考虑在内,利用代价矩阵生成一对阈值。但决策粗糙集不具备经典粗糙集的单调性,这为粗糙集的属性约简带来了新的挑战。传统的决策粗糙集中的代价矩阵只有一个,没有考虑到代价的变化性。首先介绍了多代价决策粗糙集下的悲观决策规则和乐观决策规则的定义,利用多个代价矩阵来生成阈值,并将其用于属性约简中。在属性约简中,从单独的决策类出发而不是基于全部的决策类提出了启发式的Local属性约简方法,且从相关实验结果中可以得到,相对于基于全部的决策类的属性约简,Local属性约简在乐观条件下比在悲观条件下能获得更多的正域规则。

Compared with classic rough set,traditional decision-theoretic rough set takes the cost into consideration,using cost matrix to generate a pair of thresholds.But decision-theoretic rough set doesn’t meet the monotonicity that has been widely used in classic rough set,which has brought a new challenge for us in the study of attribute reduction in rough set. Cost matrix in traditional decision-theoretic rough set is only one,doesn’t think about the variability of cost.The pessimistic decision rules and the optimistic rules of muticost decision-theoretic rough set are introduced at first and the thresholds which generated by multiple cost matrix are applied to attribute reduction.An heuristic Local attribute reduction method is proposed not on whole decision class but on individual decision class,which can get more positive rules from relevant experiment results in optimistic conditions than in pessimistic conditions,when it compared with the method based on the whole decision class.