基于FDR的证据理论改进算法  

Improved Algorithm of Evidence Theory Based on Feature Dimension Reduction

作　　者：侯庆山 邢进生[1] HOU Qing-shan;XING Jin-sheng(School of Mathematics and Computer Science,Shanxi Normal University,Linfen 041000,China)

机构地区：[1]山西师范大学数学与计算机科学学院,山西临汾041000

出　　处：《计算机技术与发展》2020年第6期59-64,共6页Computer Technology and Development

基　　金：山西省软科学基金资助项目(2011041033-03)。

摘　　要：证据理论的合成规则作为一项重要的研究课题,是样本分类及决策的关键制约因素。大多数融合方法随着特征数量的增加,特征间关联随之增强,融合过程也变得更为复杂,进而导致证据融合结果不够理想。因此,提出一种基于特征降维的证据理论改进算法,该算法主要包括两方面:首先,对原始数据集进行特征降维(feature dimension reduction,FDR),降低数据集中样本特征之间的关联性,进而优化融合结果;其次,对降维后的数据集进行特征融合,由于原始数据样本特征数量的下降,融合过程也变得更为简单,进一步通过计算样本的基本概率分配(basic probability assignment,BPA),得出样本分类结果。实验表明,基于特征降维的证据理论改进算法相较于其他融合算法,融合过程更为简单,融合效果较为良好,经过Instacart数据集测试,最终的平均类型识别率为94%。As an important research topic,the synthesis rule of evidence theory is the key constraints of sample classification and decision making.In most fusion methods,as the number of features increases,the correlation among features increases,and the fusion process becomes more complex,leading to unsatisfactory results of evidence fusion.Therefore,an improved algorithm of evidence theory based on feature dimension reduction is proposed,which mainly includes two aspects.Firstly,feature dimension reduction is performed on the original data set to reduce the correlation between the sample features in the data set,so as to optimize the fusion results.Secondly,feature fusion is carried out for the data set after dimension reduction.Due to the decline in the number of features of the original data samples,the fusion process becomes simpler.Further,the classification results of the samples are obtained by calculating the basic probability assignment of the samples.Experiments show that compared with other fusion methods,the proposed algorithm has a simpler fusion process and a better fusion effect.Through the Instacart data set test,the final average type recognition rate is 94%.

关 键 词：证据理论 组合规则 BPA 样本分类 特征降维 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]