一种基于相容块划分的动态增量式属性约简方法  

A dynamic incremental attribute reduction method based on consisitent block partitioning

作　　者：徐阳 王磊[1,2] 张义宗 王诚彪 Xu Yang;Wang Lei;Zhang Yizong;Wang Chengbiao(School of Information Engineering,Nanchang Institute of Engineering,Nanchang,330099,China;Jiangxi Province Key Laboratory of Water Information Cooperative Sensing and Intelligent Processing,School of Information Engineering,Nanchang Institute of Engineering,Nanchang,330099,China)

机构地区：[1]南昌工程学院信息工程学院,南昌330099 [2]江西省水信息协同感知与智能处理重点实验室,南昌工程学院信息工程学院,南昌330099

出　　处：《南京大学学报（自然科学版）》2023年第4期680-689,共10页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(61562061);江西省教育厅科技项目(GJJ170995)。

摘　　要：属性约简是数据挖掘、机器学习等研究领域中的一个颇为重要的预处理步骤,其效率的高低会直接影响相关任务的性能.针对已有的非增量式属性约简方法在相容块粗糙集模型中对象集发生变化时无法高效更新属性约简的问题,提出一种以区分度为启发信息的增量式属性约简方法.首先,引入相容块的概念并运用相容块对论域进行划分,在此基础上给出不完备信息系统的区分度定义;然后,详细分析对象集发生变化条件下区分度的更新机理;进一步,以区分度为启发式信息构造增量式属性约简算法;最后,选取六个UCI数据集进行增量式约简的更新实验.实验结果表明,在不影响属性约简精度的前提下,该增量式方法的时间消耗比非增量式更新方法平均缩短50%,更加可行和高效.Attribute reduction is essential for preprocessing in research fields such as data mining and machine learning since its efficiency has an immediate impact on the performances of the above⁃mentioned fields.In this paper,an incremental attribute reduction approach with discrimination as heuristic information is proposed against the problem of the non⁃incremental attribute reduction method failing to efficiently update the attribute reduction when the object set is changed in the consistent block rough set model.To begin with,the concept of compatible block is introduced to divide the domain of discourse using compatible blocks.Then,the differentiation degree of incomplete information system is defined before the analysis on the update mechanism of discrimination when the object set is changed.Moreover,an incremental attribute reduction algorithm is built with the discrimination as heuristic information.At last,an incremental reduction updating experiment is conducted on six UCI data sets selected.The time consumed by the incremental method is 50%shorter than that by the non⁃incremental updating method on the premise of not affecting the accuracy of attribute reduction,according to the experimental results.To sum up,the proposed incremental reduction algorithm is more feasible and efficient than the non⁃incremental reduction algorithm.

关 键 词：属性约简 相容块 划分 区分度 增量学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]