摘要
多标签学习的一个重要挑战是特征维度灾难.为了寻求高效的多标签特征选择方法,本文从模糊粗糙集和依赖决策熵的角度研究多标签特征选择,提出多标签特征选择新方法.首先,定义了多标签模糊信息系统,利用模糊标签粒的近似集提出了模糊决策熵和模糊依赖决策熵,研究了它们的性质.在多标签模糊信息系统上提出了基于模糊依赖决策熵的约简定义,进而给出了特征的重要性度量,以及基于模糊依赖决策熵的多标签特征选择方法和算法.最后在10个公共多标签数据集上对5种指标进行参数分析和性能对比.结果表明,所提算法具有一定的有效性,在大多数指标上优于PMU、MDDM等多标签特征选择算法.
Feature dimension disaster is one of the important challenges of multi-label learning.In order to seek an efficient multi-label feature selection method,this paper studies multi-label feature selection from the perspective of fuzzy rough set and decision dependent entropy,and proposes a new method for multi-label feature selection.Firstly,a multi-label fuzzy information system is defined,and fuzzy decision entropy and fuzzy dependent decision entropy are proposed by using the approximate set of fuzzy label particles,and their properties are studied.A simplified definition of fuzzy dependent decision entropy based on fuzzy dependent decision entropy is proposed on multi-label fuzzy information system,and then the importance measure of features is given,and a multi-label feature selection method and algorithm based on fuzzy dependent decision entropy are given.Finally,the parameters and performance comparison of five indicators are carried out on 10 public multi-label datasets.The results show that the proposed algorithm has certain effectiveness,and outperforms PMU,MDDM,and other multi-label feature selection algorithms on most indicators.
作者
陈曦
马建敏
刘权芳
CHEN Xi;MA Jianmin;LIU Quanfang(School of Science,Chang’an University,Xi’an 710064,China;Hangzhou DTWave Technology Co.,Ltd.,Hangzhou 311100,China)
出处
《昆明理工大学学报（自然科学版）》
北大核心
2024年第2期62-72,共11页
Journal of Kunming University of Science and Technology(Natural Science)
基金
国家自然科学基金项目(61772019)
科技部国家重点研发计划项目(2022YFC3303100)。
关键词
多标签学习
特征选择
模糊粗糙集
模糊决策熵
模糊依赖决策熵
multi-label learning
feature selection
fuzzy rough sets
fuzzy decision entropy
fuzzy dependent decision entropy
