摘要
应用特征选择处理多标签数据分类时"维度灾难"问题已成为重要研究方向,因此提出一种基于邻域维护准则的特征选择算法(NPFS,feature selection algorithm based on neighborhood preservation criterion)。通过近似基于特征子空间和基于标签空间的2个相似度矩阵来构建相似性维护表达式,再通过线性近似扩展相似性维护公式得到邻域关系维护公式,并计算出邻域关系维护得分(NRPS,neighborhood relationship preserving score)来评估特征子集的重要性,结合贪婪方法设计具有NRPS的多标签特征选择算法(NPFS)。仿真结果表明,对比MMIFS算法和MDMR算法,所提出的算法在平均准确率、覆盖率、汉明损失、1-错误率、排名损失5个性能指标上均有改善。
The application of feature selection to deal with'Dimensional Disaster'in multi-label data classification has become an important research direction,we proposed a feature selection algorithm based on neighborhood preservation criterion(NPFS).A similarity preservation expression was constructed by approximating two similarity matrices based on feature subspace and label space.Then,the similarity preservation formulation was extended by the linear approximation to obtain a formulation of the neighborhood relationship preservation,and the importance of the feature subset was evaluated by calculating the neighborhood relationship preserving score(NRPS).A multi-label feature selection algorithm with NRPS was designed in combination with the greedy method(NPFS).The simulation results show that the metrics of average precision,coverage,hamming loss,one-error,ranking loss obtained by the proposed algorithm have been improved compared with those obtained by MMIFS algorithm and MDMR algorithm.
作者
刘云
肖雪
LIU Yun;XIAO Xue(School of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,P.R.China)
出处
《重庆大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2019年第3期58-64,共7页
Journal of Chongqing University
基金
国家自然科学基金资助项目(61262040)~~
关键词
特征选择
多标签分类
邻域关系维护得分
贪婪算法
feature selection
multi-label classification
neighborhood relationship preserving score
greedy algorithm
