基于邻域决策误差率的层次分类在线流特征选择  

Online Hierarchical Streaming Feature Selection Based on Neighborhood Decision Error Rate

作　　者：王晨曦 刘园奎 吕彦 林耀进 Wang Chenxi;Liu Yuankui;LüYan;Lin Yaojin(School of Computer Science,Minnan Normal University,Zhangzhou 363000,China;Key Laboratory of Data Science and Intelligence Application,Minnan Normal University,Zhangzhou 363000,China)

机构地区：[1]闽南师范大学计算机学院,福建漳州363000 [2]闽南师范大学数据科学与智能应用福建省高等学校重点实验室,福建漳州363000

出　　处：《南京师范大学学报（工程技术版）》2022年第4期9-18,共10页Journal of Nanjing Normal University(Engineering and Technology Edition)

基　　金：国家自然科学基金项目(62076116);福建省自然科学基金项目(2020J01811、2020J01792和2021J02049).

摘　　要：在实际应用领域中,存在许多特征空间无法预先给定的场景,数据以特征流的形式随时间动态流入特征空间,而样本数量是固定不变的.同时,数据的类别中往往存在丰富的层次化结构关系,传统的特征选择算法在性能上已无法满足需求.基于此,本文提出一种面向层次分类学习的在线流特征选择算法.首先,利用兄弟节点之间的关系设计了一种基于最大近邻的决策误差率计算公式.其次,设计在线重要性选择和在线冗余更新两种在线评估准则,用于选择决策误差最小的特征子集.最后,在6个层次数据集上的实验结果表明,所提算法优于一些现有的在线流特征选择算法.In many practical application fields,there are numerous scenes in which the entire feature space cannot be available in advance,candidate features flow into the feature space dynamically over time,and the number of samples is fixed.At the same time,there exists a hierarchical structure relationship between classes,and traditional feature selection methods cannot be able to meet the demand.Based on these,an online streaming feature selection algorithm for hierarchical classification learning is presented.Firstly,a decision error rate calculation formula is designed on the basis of the largest nearest neighbor according to sibling relationships.Secondly,two online evaluation criteria of online significance selection and online relevance analysis are proposed to select features with minimum decision error.Finally,experimental results on six hierarchical datasets manifest that the proposed algorithm is better than some existing online streaming feature selection algorithms.

关 键 词：在线流特征选择 层次分类 兄弟关系 邻域决策误差率 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]