Dempster-Shafer证据理论驱动的邻域粗糙分类方法

A Neighborhood Rough Classification with Dempster-Shafer Evidence Theory

为了进一步改进邻域分类器的分类机制,提升分类性能,提出Dempster-Shafer(D-S)证据理论驱动的邻域粗糙分类方法。首先,采用邻域决策错误率作为属性重要性的指标研究基于邻域决策错误率的属性约简方法,通过删除冗余属性,为分类学习提供重要的属性集合;其次,改变传统多数投票机制,将D-S证据理论引入邻域样本的信息融合中,提出基于D-S证据理论的邻域分类器;最后,基于UCI公共数据集的实验结果表明,所提方法相较于多数投票机制下的邻域分类器,具有更高的分类精度,为邻域分类方法的进一步研究提供了新的思路。

In order to further improve the classification mechanism and the performance of neighborhood classifier,a Dempster-Shafer(D-S)evidence-driven neighborhood rough classification method is proposed.Firstly,in attribute reduction,the error rate of neighborhood decision is used as the index of attribute significance,and the attribute reduction method based on neighborhood decision error rate is studied.By removing redundant attributes,an important set of attributes is provided for classification learning.Then,in terms of classifier design,the traditional majority voting mechanism is revised,D-S evidence theory is introduced into the information fusion of neighborhood samples,and a neighborhood classifier based on D-S evidence theory is proposed.Experimental results on UCI public data set show that the proposed method has higher classification accuracy than the neighborhood classifier under the majority voting mechanism.The paper provides a new insight for the further study of neighborhood classification methods.