基于特征选择和模糊类支持度的模糊分类关联规则挖掘算法  

Fuzzy Classification Association Rule Mining Algorithm Based on Feature Selection and Fuzzy Category Support

作　　者：王子恒 李鹏[1,2] 陈静[1] WANG Ziheng;LI Peng;CHEN Jing(School of Computer Science,Nanjing University of Posts and Telecommunications,Nanjing Jiangsu 210023;Jiangsu High Technology Research Key Laboratory for Wireless Sensor Networks,Jiangsu Province,Nanjing Jiangsu 210023)

机构地区：[1]南京邮电大学计算机学院,江苏南京210023 [2]江苏省无线传感网高技术研究重点实验室,江苏南京210023

出　　处：《软件》2023年第8期15-22,共8页Software

基　　金：国家自然科学基金(62102194);江苏省六大人才高峰高层次人才项目(RJFW-111)资助

摘　　要：模糊分类关联规则(Fuzzy Classification Association Rules,FCAR)是一种特殊的模糊关联规则,挖掘FCAR对于构建基于规则的分类模型至关重要。传统关联规则挖掘算法挖掘FCAR时可能会包含较多冗余规则,并且在数据集类别不平衡时,挖掘到的小类规则的数量会急剧减少甚至降为0。为解决上述问题,提出了一种基于特征选择和模糊类支持度-模糊提升度框架(Fuzzy Category Support-Fuzzy Lift Framework,FCS-FLF)的FCAR挖掘算法FSFCS Based FCARMiner(Feature Selection and Fuzzy Category Support-Fuzzy Lift Framework Based FCAR-Miner),基于模糊隶属度矩阵迭代挖掘FCAR。在多个类别不平衡的数据集上的实验结果表明,相比其他算法FSFCS Based FCAR-Miner算法能够避免大量冗余规则的生成,同时也能适应数据类别不平衡的情况,不会出现各类规则数量相差悬殊的情况。Fuzzy Classification Association Rules(FCAR)is a special type of fuzzy association rule that plays a crucial role in constructing rule-based classification models.Traditional association rule mining algorithms may generate a large number of redundant rules when mining FCAR,and when dealing with imbalanced datasets,the number of rules discovered for minority classes can drastically decrease or even become zero.To address these issues,a FCAR mining algorithm called FSFCS based FCAR-Miner(Feature Selection and Fuzzy Category Support-Fuzzy Lift Framework Based FCAR-Miner)is proposed.It utilizes a fuzzy category support-fuzzy lift framework and incorporates feature selection techniques to iteratively mine FCAR based on a fuzzy membership degree matrix.Experimental results on multiple imbalanced datasets demonstrate that compared to other algorithms,FSFCS Based FCAR-Miner can avoid generating a large number of redundant rules and also handle imbalanced data classes without significant discrepancies in the number of rules discovered for each class.

关 键 词：模糊分类关联规则挖掘 特征选择 类别不平衡数据 模糊类支持度 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]