基于eEP的两阶段分类方法的研究  

Based on essential emerging patterns to classify in two phases

作　　者：职为梅[1] 周文刚[2] 孙宜贵[3] 

机构地区：[1]郑州大学计算机科学系,河南郑州450052 [2]北方工业大学计算机科学系,北京100041 [3]河南工业大学计算机科学系,河南郑州450052

出　　处：《商丘师范学院学报》2006年第2期88-92,共5页Journal of Shangqiu Normal University

摘　　要：传统的基于规则的分类算法多是采用顺序覆盖技术训练分类规则,这使得训练得到的模型覆盖大量的非目标类实例,分类时效果差.基于规则的两阶段分类算法,能够很好的去除模型覆盖的非目标类实例,分类时能取得比较好的结果.EP在分类大型数据库时能够提高分类效率,eEP(Essential Emerging Patterns)是一种特殊的EP,较EP能够减少分类噪音.本文中我们构造一个新颖的分类算法,基于eEP的两阶段分类方法(即EEPCTP),并使用UCI机器学习库中的10个数据集做实验,实验表明EEPCTP分类法取得了与一些经典的分类算法可比的效率和准确性.Traditional algorithms for classification usually use technique of sequential covering to train model, but the methods can make the classifier cover many examples of the non-target class and affects the accuracy. Two Phase to Classify, which can remove most of the exemples of the non-target class that were covered by the classifier, can receive a good effect when classify rare class. EP is good for classification of large database, eEP is special EP which can reduce the noise of classification. In the paper we propose a novel approach, EEPCTP, which can be looked as a hybrid of eEP-based classifier and Two Phase classifier and use UCI Machine Leaming Repository as experimental dataset. We prove that our method can achieve good effect.

关 键 词：数据挖掘 分类 显露模式 两阶段分类 

分 类 号：TP311.12[自动化与计算机技术—计算机软件与理论]