邻域覆盖分类的两种加权策略  

Two weighted strategies for neighborhood covering classification

作　　者：王轩 刘福伦[1] 张林[1] 王宏杰[1] 闵帆[1] WANG Xuan;LIU Fulun;ZHANG Lin;WANG Hongjie;MIN Fan(School of Computer Science,Southwest Petroleum University,Chengdu Sichuan 610500,China)

机构地区：[1]西南石油大学计算机科学学院,成都610500

出　　处：《计算机应用》2018年第A01期1-5,19,共6页journal of Computer Applications

基　　金：国家自然科学基金资助项目(61379089)

摘　　要：邻域覆盖粗糙集在机器学习的理论与应用中都起着重要作用。结合覆盖约简和代表选择,已有的研究在符号数据上取得了很好的分类结果;然而,已有方法使用简单投票策略,不能有效解决分类冲突问题。对此提出两种新的加权策略,在分类阶段确定投票的模式。第一种基于Cfs Subset EVal和Best First根据属性重要度加权,第二种基于预测点与代表的Overlap相似度加权。利用加州大学欧文分校(UCI)的10个公开数据集进行实验,并与其他三种常用分类算法进行对比。实验用F-measure值对算法性能进行评定。实验结果表明,两种新策略均能提升分类精度,其中属性加权策略效果更明显。此外,属性加权策略特别适用于对生命领域数据集进行分类。Neighborhood covering rough set plays an important role in the theory and application of machine learning. Combining covering reduction and representative selection, existing studies have obtained good classification results on symbolic data. However, the existing methods use simple voting strategy, and can not solve the problem of classification conflict effectively. Two new weighting strategies were proposed to determine the voting pattern in the classification stage. The first one was weighting by the significance of attributes, using the CfsSubsetEVal algorithm and the BestFirst algorithm, and the second one was weighted by the overlap similarity between the prediction points and the representative points. Experiments were conducted on 10 public datasets of University of Califomialrvine （UCI） and compared with three other commonly used classification algorithms. The performance of the proposed algorithms were evaluated by F-measure values. The experimental results indicate that both the two new strategies can improve the classification accuracy, among which the attribute weighting strategy is more effective. In addition, attribute weighting strategy is particularly suitable for classifying life domain datasets.

关 键 词：代表选择 邻域 相似度 覆盖约简 分类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]