基于归一化互信息的FCBF特征选择算法  

FCBF algorithm based on normalized mutual information for feature selection

作　　者：段宏湘[1] 张秋余[1] 张墨逸[1] 

机构地区：[1]兰州理工大学计算机与通信学院,甘肃兰州730050

出　　处：《华中科技大学学报（自然科学版）》2017年第1期52-56,共5页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61363078);甘肃省青年科技基金资助项目(148RJYA001)

摘　　要：针对高维数据中不相关特征、冗余特征等导致的分类任务计算量大、分类正确率低等问题,提出了一种基于归一化互信息的相关性快速过滤特征选择(FCBF-NMI)算法.该算法采用归一化互信息代替对称不确定性作为FCBF算法的相关性评价标准,进行特征与类别、特征与特征的相关性分析,删除不相关特征及冗余特征以获得最优特征子集.实验结果表明:FCBF-NMI算法得到的最优特征子集更合理,平均分类正确率为89.68%,所用时间平均低至2.64s.As the irrelevant and redundant features of the high-dimensional data can cause the large computational quality and low accuracy of classification,a FCBF feature selection algorithm based on normalized mutual information was proposed,named FCBF-NMI.Firstly,by replacing symmetric uncertainty with the normalized mutual information,the correlation evaluation standards of the FCBF algorithm were established.Secondly,the standards could be used to analyze the correlation between features and the class or features.Finally,by the removing of the irrelevant and redundant features,the optimal feature subset could be obtained.The experiments results show that the optimal feature subset achieved by FCBF-NMI is more reasonable,and the mean classification accuracy rate can reach to 89.68%,and the average time is 2.64 s.

关 键 词：高维数据 特征选择 归一化互信息 相关性快速过滤特征选择(FCBF) 分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]