动态邻域粒化方式下的属性约简研究  

Research on Attribute Reduction Via Dynamic Neighborhood Granulation Method

作　　者：邓安生[1] 赵梓旭 DENG An-sheng;ZHAO Zi-xu(Information Science and TechnologyCollge,Dalian Maritime University,DalianLiaoning116026,China)

机构地区：[1]大连海事大学信息科学技术学院,辽宁大连116026

出　　处：《计算机仿真》2023年第8期327-333,共7页Computer Simulation

摘　　要：在邻域粗糙集研究中,往往通过给定的半径来约束样本之间的相似性进而实现邻域信息粒化。若在粒化过程中给定的半径较大,导致不同类别的样本将落入同一邻域中,则会引起信息粒化不精确和不一致。为解决上述问题,考虑到总体样本分布和在约简过程中属性和标签之间的关系对选择邻域半径的影响,提出一种动态邻域的信息粒化机制,然后构造了动态邻域粗糙集模型。在此基础上提出一种多准则属性评估方法,使用前向贪心搜索策略实现了约简求解。在12个公共数据集上进行仿真测试,结果表明,与传统的邻域粗糙集等得到的约简相比,所提方法对不同的数据集可选择合适数量的属性,且提供更高的分类精度。In the research on neighborhood rough set,the similarity between samples is often restricted by a given radius to achieve neighborhood information granulation.If the given radius is large in the granulation process,samples of different types will fall into the same neighborhood,which will cause inaccurate and inconsistent information granulation.In order to solve this problem,a dynamic neighborhood information granulation mechanism is proposed.In order to solve this problem,considering the influence of the overall sample distribution and the relationship between attributes and labels in the reduction process on the size of the radius of the selected neighborhood,a dynamic neighborhood information granulation mechanism was proposed.First,the dymanic neighborhood rough set model was constructed.On this basis,a multi-criteria attribute evaluation method is proposed,which uses forward greedy search strategy to achieve reduction solution.The simulation test was conducted on 12 public data sets,and the results show that,compared with the reduction obtained by traditional methods such as neighborhood rough set,the proposed method can select an appropriate number of attributes for different data sets and provide higher classification accuracy.

关 键 词：信息粒化 动态邻域 粗糙集 多准则属性约简 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]