WilsonTh数据剪辑在邻域粗糙协同分类中的应用  

Application of WilsonTh Data Editing for Neighborhood Rough Sets Based Co-training Classification Model

作　　者：张维[1,2,3] 苗夺谦[1,3] 李峰[1,3] 

机构地区：[1]同济大学电子与信息工程学院,上海201804 [2]上海电力学院计算机科学与技术学院,上海200090 [3]同济大学嵌入式系统与服务计算教育部重点实验室,上海201804

出　　处：《计算机科学与探索》2014年第9期1092-1100,共9页Journal of Frontiers of Computer Science and Technology

基　　金：国家自然科学基金;高等学校博士学科点专项科研基金~~

摘　　要：邻域粗糙协同分类模型结合了邻域粗糙集和协同学习理论,可以处理连续型数据,并可有效利用无标记数据提高分类的性能。但在学习过程中,无标记数据常被错误地标记,从而给训练集引入噪声数据,并导致分类性能不稳定。针对该问题,探讨了WilsonTh数据剪辑在邻域粗糙协同分类模型中的应用。在每一次迭代学习过程中,分类器给无标记数据加上类别标记后,应用WilsonTh数据剪辑选出最大可能标记正确的样本加入训练集,分类器在扩大的训练集上再训练以获得更好的性能。UCI数据集上实验结果表明,WilsonTh数据剪辑能有效地提高加入训练集的数据质量,从而增强邻域粗糙协同分类的性能。A neighborhood rough sets based co-training classification model can deal with continuous data and utilize the unlabeled and labeled data to achieve better performance than the classifiers learning only from few labeled data. However, in the learning process, the unlabeled data may be wrongly labeled, which would introduce the noise to the training set and result in the instability of classification performance. This paper discusses the application of WilsonTh data editing in the neighborhood rough sets based co-training classification model. In the iteration, the classifiers label the unlabeled data with class symbol, use WilsonTh data editing to select right labeled data to the greatest extent pos-sible, and add these labeled data to enlarge the training set for better quality of classifiers retraining. The experimental results on selected UCI datasets show that the application of WilsonTh data editing is more effective to improve the learning accuracy of neighborhood rough sets based co-training classification model.

关 键 词：WilsonTh数据剪辑 邻域粗糙集 邻域互信息 协同学习 连续型数据 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]