模糊粗糙集的LIFT数据分析  

LIFT Data Analysis by Fuzzy Rough Sets

作　　者：梁绍宸 徐苏平[1] 窦慧莉[1] 李洪梅[1] 杨习贝[1,2] LIANG Shao-chen;XU Su-ping;DOU Hui-lil;LI Hong-mei;YANG Xi-bei(School of Computer,Jiangsu University of Science and Teclmology,Zhenjiang 212003 ,China;School of Economics ＆ Management, Nanjing University of Science and Technology,Nanjing 210094, China)

机构地区：[1]江苏科技大学计算机学院,江苏镇江212003 [2]南京理工大学经济管理学院,南京210094

出　　处：《小型微型计算机系统》2018年第5期1052-1057,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61572242,61503160,61502211)资助; 江苏省高校哲学社会研究一般项目(2015SJD769)资助; 中国博士后科学基金项目(2014M550293)资助

摘　　要：在多标记学习中,因为不同的标记拥有与其自身紧密相关的特性,所以可以利用LIFT策略来处理多标记问题,其过程包含两个步骤:首先根据不同标记构建类属属性,然后在类属属性空间上进行分类.然而由于利用LIFT所构建的类属属性维度较高,会致使分类模型训练变慢或泛化能力不足.为解决这一问题,借助传统与稳健的模糊粗糙集模型,提出了对类属属性空间进行特征选择,并在此基础上利用模糊粗糙分类器进行多标记预测的模糊粗糙LIFT方法.实验结果表明,新算法不仅可以有效地降低类属属性空间维度,而且在压缩后的类属属性空间中,分类性能将有所提升.In multi-label learning,since different labels may have different characteristics,a strategy called LIFT can be used to deal with multi-label. LIFT contains two steps： firstly,it constructs label-specific features for different labels; secondly,it performs classification in label-specific feature spaces. However,the dimensionality of label-specific features constructed from LIFT is too high to be good at training classification models or advancing generalization ability. To solve this problem,through using classical and robust fuzzy rough set models,FRS-LIFT is proposed,which involves feature selection in label-specific feature spaces and multi-label predictions based on fuzzy rough set classifiers. The experimental results show that the new algorithm can not only effectively reduce the dimensionalities of label-specific feature spaces,but also improve the predictive performance.

关 键 词：类属属性 多标记学习 模糊粗糙集 特征选择 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]