基于Fisher Score和社会选择优化的特征选择方法  

A novel feature selection approach with Fisher Score and social selection optimization

作　　者：郑艺峰 李国和[3,4] 张文杰 潘雪玲 魏葆雅 ZHENG Yifeng;LI Guohe;ZHANG Wenjie;PAN Xueling;WEI Baoya(School of Computer Science,Minnan Normal University,Zhangzhou,Fujian 363000,China;Key Laboratory of Data Science and Intelligence Application,Fujian Province University,Zhangzhou,Fujian,363000,China;Beijing Key Lab of Petroleum Data Mining,China University of Petroleum-Beijing,Beijing 102249,China;College of Information Science and Engineering,China University of Petroleum,Beijing 102249,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 [3]中国石油大学石油数据挖掘北京市重点实验室,北京102249 [4]中国石油大学信息科学与工程学院,北京102249

出　　处：《闽南师范大学学报（自然科学版）》2021年第1期72-79,共8页Journal of Minnan Normal University：Natural Science

基　　金：福建省自然科学基金(2018J01546,2019J01748);福建省教育厅中青年项目(JAT190392);克拉玛依科技发展计划项目(2020CGZH0009);中国石油大学(北京)克拉玛依校区科研基金(RCYJ2016B-03-001)。

摘　　要：随着信息技术的发展,智能应用已经深入社会生活的各个方面.与此同时,数据量急剧增加,容易导致"维度灾难",从而降低分析模型的性能.为此,本文提出基于社会选择理论的特征选择方法,简称FBMN.首先,将社会选择理论与Fisher Score相结合用于计算特征相关性,从而获得稳定的近似全局最优特征排序.其次,提出平均不确定信息度量,并结合马尔科夫毯以有效的剔除冗余特征.最后,公开数据集的实验结果表明,FBMN有助于提高分类模型的性能,尤其在高维数据.With the development of information technology,intelligent applications have penetrated all aspects of social life.Simultaneously,the increase of data will lead to the curse of dimensionality,thereby reducing the performance analysis model.In this paper,we propose a feature selection method based on social selection theory,called FBMN.First,the social choice theory and Fisher Score are combined to calculate feature correlation to obtain a stable approximate global optimal feature ranking.Then,a Markov blanket with the average uncertain information is utilized to eliminate redundant features effectively.Eventually,extensive experiments results on publicly data sets show that FBMN can improve the performance of the classification model,especially in high-dimensional data.

关 键 词：数据挖掘 特征选择 社会选择理论 信息不确定性 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]