混合型数据的邻域条件互信息熵属性约简算法

doi:10.11959/j.issn.2096-0271.2022066

摘要/Abstract

摘要：

属性约简是粗糙集理论的重要研究内容之一，其主要目的是消除信息系统中不相关的属性，降低数据维度并提高数据知识发现性能。然而，基于粗糙集的属性约简方法大多没有考虑属性之间的依赖性，使得最终的属性约简结果存在一定的冗余属性。对此，提出一种基于邻域条件互信息熵的属性约简算法。首先，在传统邻域熵的基础上，针对混合型数据，提出混合型邻域互信息熵模型和混合型邻域条件互信息熵模型；然后利用这两种熵模型进行混合型信息系统的属性依赖度评估和属性启发式搜索，并设计出一种属性约简算法；最后通过UCI数据集的实验分析，证明了提出的算法具有较高的属性约简性能。

关键词: 粗糙集, 属性约简, 邻域, 互信息熵, 条件互信息熵

Abstract:

Attribute reduction is an important research content of the rough set theory.Its main purpose is to eliminate irrelevant attributes in information systems, reduce data dimensions and improve data knowledge discovery performance.However, most of the attribute reduction methods based on a rough set do not consider the dependence between attributes, which makes the final attribute reduction result have some redundant attributes.An attribute reduction algorithm based on neighborhood conditional mutual information entropy was proposed.Firstly, based on the traditional neighborhood entropy, a hybrid neighborhood mutual information entropy model and a hybrid neighborhood conditional mutual information entropy model were proposed for hybrid data.Then, the two entropy models were used to evaluate the attribute dependence and attribute heuristic search of the hybrid information system, and an attribute reduction algorithm was designed.Finally, through the experimental analysis of UCI data sets, it was proved that the algorithm had higher attribute reduction performance.

Key words: rough set, attribute reduction, neighborhood, mutual information entropy, conditional mutual information entropy

中图分类号:

TP18

兰海波. 混合型数据的邻域条件互信息熵属性约简算法[J]. 大数据, 2022, 8(4): 133-144.

Haibo LAN. Neighborhood conditional mutual information entropy attribute reduction algorithm for hybrid data[J]. Big Data Research, 2022, 8(4): 133-144.

图/表 8

表1

表2

表3

表4

图1

图2

图3

图4

参考文献 20

[1]	YANG X , LI M M , FUJITA H ,et al. Incremental rough reduction with stable attribute group[J]. Information Sciences, 2022,589: 283-299.
[2]	SUN L , YIN T Y , DING W P ,et al. Multilabel feature selection using MLReliefF and neighborhood mutual information for multilabel neighborhood decision systems[J]. Information Sciences, 2020,537: 401-424.
[3]	周涛, 陆惠玲, 任海玲 ,等. 基于粗糙集的属性约简算法综述[J]. 电子学报, 2021,49(7): 1439-1449.
	ZHOU T , LU H L , REN H L ,et al. Survey on attribute reduction algorithm of rough set[J]. Acta Electronica Sinica, 2021,49(7): 1439-1449.
[4]	HU Q H , YU D R , LIU J F ,et al. Neighborhood rough set based heterogeneous feature subset selection[J]. Information Sciences, 2008,178(18): 3577-3594.
[5]	PANG Q Q , ZHANG L . Semi-supervised neighborhood discrimination index for feature selection[J]. Knowledge-Based Systems, 2020,204: 106224.
[6]	HU M , TSANG E C C , GUO Y T ,et al. A novel approach to attribute reduction based on weighted neighborhood rough sets[J]. Knowledge-Based Systems, 2021,220: 106908.
[7]	SHU W H , QIAN W B , XIE Y H . Incremental feature selection for dynamic hybrid data using neighborhood rough set[J]. Knowledge-Based Systems, 2020,194: 105516.
[8]	盛魁, 王伟, 卞显福 ,等. 混合数据的邻域区分度增量式属性约简算法[J]. 电子学报, 2020,48(4): 682-696.
	SHENG K , WANG W , BIAN X F ,et al. Neighborhood discernibility degree incremental attribute reduction algorithm for mixed data[J]. Acta Electronica Sinica, 2020,48(4): 682-696.
[9]	姚晟, 李初宴, 陈悦 . 基于非平衡数据下不完备混合型信息系统的属性约简[J]. 计算机应用研究, 2021,38(5): 1331-1335.
	YAO S , LI C Y , CHEN Y . Attribute reduction of incomplete hybrid information system based on unbalanced data[J]. Application Research of Computers, 2021,38(5): 1331-1335.
[10]	WANG C Z , HUANG Y , DING W P ,et al. Attribute reduction with fuzzy rough self-information measures[J]. Information Sciences, 2021,549: 68-86.
[11]	YUAN Z , CHEN H M , LI T R ,et al. Unsupervised attribute reduction for mixed data based on fuzzy rough sets[J]. Information Sciences, 2021,572: 67-87.
[12]	栾雨雨, 王锡淮, 肖健梅 . 基于混沌离散粒子群的粗糙集属性约简算法[J]. 计算机仿真, 2021,38(7): 271-275.
	LUAN Y Y , WANG X H , XIAO J M . Rough set attribute reduction algorithm based on chaotic discrete particle swarm optimization[J]. Computer Simulation, 2021,38(7): 271-275.
[13]	HU M , TSANG E C C , GUO Y T ,et al. Attribute reduction based on overlap degree and k-nearest-neighbor rough sets in decision information systems[J]. Information Sciences, 2022,584: 301-324.
[14]	桑彬彬, 杨留中, 陈红梅 ,等. 优势关系粗糙集增量属性约简算法[J]. 计算机科学, 2020,47(8): 137-143.
	SANG B B , YANG L Z , CHEN H M ,et al. Incremental attribute reduction algorithm in dominance-based rough set[J]. Computer Science, 2020,47(8): 137-143.
[15]	熊菊霞, 吴尽昭, 王秋红 . 邻域互信息熵的混合型数据决策代价属性约简[J]. 小型微型计算机系统, 2021,42(8): 1584-1590.
	XIONG J X , WU J Z , WANG Q H . Decision cost attribute reduction of hybrid data based on neighborhood mutual information entropy[J]. Journal of Chinese Computer Systems, 2021,42(8): 1584-1590.
[16]	陈帅, 张贤勇, 唐玲玉 ,等. 邻域互补信息度量及其启发式属性约简[J]. 数据采集与处理, 2020,35(4): 630-641.
	CHEN S , ZHANG X Y , TANG L Y ,et al. Neighborhood complementary information measures and heuristic attribute reduction[J]. Journal of Data Acquisition and Processing, 2020,35(4): 630-641.
[17]	陈帅 . 基于三层粒结构的邻域互补信息度量及其属性约简[D]. 成都:四川师范大学, 2020.
	CHEN S . Neighborhood complementary information measures and their attribute reductions based on three-layer granular structure[D]. Chengdu:Sichuan Normal University, 2020.
[18]	姚晟, 徐风, 吴照玉 ,等. 基于邻域粗糙互信息熵的非单调性属性约简[J]. 控制与决策, 2019,34(2): 353-361.
	YAO S , XU F , WU Z Y ,et al. Nonmonotonic attribute reduction based on neighborhood rough mutual information entropy[J]. Control and Decision, 2019,34(2): 353-361.
[19]	SALEM O A M , LIU F , CHEN Y P P ,et al. Feature selection and threshold method based on fuzzy joint mutual information[J]. International Journal of Approximate Reasoning, 2021,132: 107-126.
[20]	HU Q H , ZHANG L , ZHANG D ,et al. Measuring relevance between discrete and continuous features based on neighborhood mutual information[J]. Expert Systems With Applications, 2011,38(9): 10737-10750.

名称	样本数/个	属性数/个	分类数目/个
Cylinder	512	40	3
Credit	690	15	2
German	1 000	19	2
Segment	2 310	19	7
Sick	2 800	28	3
Abalone	4 177	8	29

对比项	原始数据集	对比算法1^[6]	对比算法2^[10]	对比算法3^[19]	本文算法
Cylinder	0.81±0.018	0.86±0.011	0.83±0.024	0.86±0.013	0.89±0.009
Credit	0.72±0.023	0.71±0.008	0.77±0.021	0.76±0.013	0.80±0.013
German	0.73±0.005	0.82±0.014	0.85±0.031	0.86±0.018	0.84±0.016
Segment	0.82±0.009	0.83±0.018	0.86±0.027	0.84±0.017	0.88±0.017
Sick	0.86±0.021	0.93±0.020	0.91±0.021	0.90±0.012	0.95±0.011
Abalone	0.82±0.040	0.89±0.014	0.86±0.043	0.83±0.025	0.89±0.016
平均	0.79±0.019	0.84±0.014	0.85±0.028	0.83±0.016	0.88±0.014

对比项	原始数据集	对比算法1^[6]	对比算法2^[10]	对比算法3^[19]	本文算法
Cylinder	0.80±0.021	0.83±0.015	0.82±0.022	0.85±0.011	0.88±0.012
Credit	0.74±0.017	0.77±0.014	0.79±0.026	0.83±0.015	0.81±0.018
German	0.74±0.009	0.86±0.018	0.83±0.027	0.84±0.012	0.87±0.014
Segment	0.80±0.012	0.82±0.015	0.84±0.015	0.83±0.018	0.86±0.016
Sick	0.83±0.014	0.90±0.021	0.88±0.017	0.89±0.014	0.93±0.013
Abalone	0.78±0.032	0.87±0.016	0.84±0.026	0.82±0.019	0.87±0.019
平均	0.78±0.018	0.84±0.016	0.83±0.022	0.85±0.015	0.87±0.015