基于集成学习的决策粗糙集特定类属性约简算法  

CLASS-SPECIFIC ATTRIBUTE REDUCTION ALGORITHM FOR DECISION-THEORETIC ROUGH SETS BASED ON ENSEMBLE LEARNING

作　　者：李明 甘秀娜 王月波 Li Ming;Gan Xiuna;Wang Yuebo(Department of Economic Management,Shijiazhuang Tiedao University Sifang College,Shijiazhuang 051132,Hebei,China;Organization and Personal Department,Shijiazhuang Institute of Railway Technology,Shijiazhuang 050041,Hebei,China;Information Technology Department,Bank of Hebei Co.,Ltd.,Shijiazhuang 050000,Hebei,China)

机构地区：[1]石家庄铁道大学四方学院经济管理系,河北石家庄051132 [2]石家庄铁路职业技术学院组织人事部,河北石家庄050041 [3]河北银行股份有限公司信息技术部,河北石家庄050000

出　　处：《计算机应用与软件》2021年第6期262-270,共9页Computer Applications and Software

基　　金：河北省文化艺术科学规划课题(HB16-YB0164)。

摘　　要：属性约简是粗糙集理论的重要研究内容。目前决策粗糙集的属性约简大多基于全局的决策类,并且都是采用单一的约简准则。针对这一问题,在决策粗糙集下提出一种特定类别属性约简算法。针对特定的决策类,给出一种属性约简的定义,在保证决策区域极大化的同时尽可能地降低决策区域划分时的代价;利用集成学习的方法设计出相应的启发式属性约简算法。通过在UCI数据集上与已有的算法进行实验比较,验证了该算法具有更高的属性约简性能。Attribute reduction is an important research content for rough set theory.At present,attribute reduction of decision-theoretic rough sets is mostly based on global decision classes,and all of them adopt a single reduction criterion.To solve this problem,this paper proposes a class-specific attribute reduction algorithm under decision-theoretic rough sets.A definition of attribute reduction was given for specific decision classes.This attribute reduction ensured the maximization of decision regions and reduced the cost of decision region division as much as possible.Then,a heuristic attribute reduction algorithm was designed by using ensemble learning method.By comparing with the existing algorithms on UCI datasets,it is proved that the proposed algorithm has higher performance of attribute reduction.

关 键 词：粗糙集 属性约简 特定类 决策区域 决策代价 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]