区间值序决策表的条件熵属性约简  

Attribute Reduction Based on Conditional Entropy in Interval Valued Ordered Decision Table

作　　者：张晓燕[1] 匡洪毅 ZHANG Xiaoyan;KUANG Hongyi(College of Artificial Intelligence,Southwest University,Chongqing 400715,China)

机构地区：[1]西南大学人工智能学院,重庆400715

出　　处：《山西大学学报（自然科学版）》2023年第1期101-107,共7页Journal of Shanxi University(Natural Science Edition)

基　　金：国家自然科学基金(61976245)。

摘　　要：由于数据自身的不确定性和观测条件有限,现实问题中许多数据以区间值形式呈现。其中,优势关系下的区间值信息表研究对于多属性决策问题有重要意义。目前针对该系统的属性约简方法主要是辨识矩阵法或基于互信息的增量式约简,但前者计算效率较低,而后者没有利用到决策信息。文章探讨了条件熵作为不确定性度量在该系统下的性质,通过比较不同属性缺失时信息系统的条件熵变化,引入了属性重要度概念,基于此提出启发式属性约简算法。最后,通过对比实验验证了本算法具有低冗余的特点,在约简率上比基于粗糙熵和正域不变等序信息系统的启发式约简。Because of the uncertainty of the data and the conditions of limited observation, many practical problems are presented as interval valued data. Among them, the study of interval-valued information system based on the dominance relationship has important significance for multi-attribute decision-making. At present, the main attribute reduction methods for this system are discernibility matrix method or incremental reduction based on mutual information, but the former is computationally inefficient, while the latter ignores decision information. This paper discusses the properties of conditional entropy as a measure of uncertainty in this system. By comparing the changes of conditional entropy of information system when different attributes are missing, the concept of attribute importance is introduced, and a heuristic attribute reduction algorithm is proposed based on this. Finally, the comparative experiment verifies that the algorithm has the characteristics of low redundancy, and the reduction rate is 14%-25% higher than that of heuristic reduction based on rough entropy and positive region-based in ordered information system.

关 键 词：区间值序决策表 条件熵 属性约简 属性重要度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]