EPTPC:一种改进的两阶段分类算法  

EPTPC:an Improved Two-Phase Classification Algorithm

作　　者：孙宜贵[1] 职为梅[2] 

机构地区：[1]河南工业大学信息科学与工程学院,河南郑州450001 [2]郑州大学信息工程学院,河南郑州450001

出　　处：《信息技术与信息化》2011年第3期73-76,共4页Information Technology and Informatization

基　　金：河南省科技攻关项目(092102210398)

摘　　要：分类是数据挖掘中的重要任务之一。基本显露模式(eEPs)是一种有趣的知识模式,能够反映两个不同数据集之间的某些显著差异并减少分类噪音,适合解决针对稠密数据集和高维数据集的分类问题。但是,传统的采用顺序覆盖策略分类算法无法解决小覆盖问题和反例碎片问题。提出了一种改进的基于eEPs模式的两阶段分类算法,它将eEPs模式作为分类模式,采用两阶段思想来构造分类器,特别是优化了评分策略和两个阶段的权重设置。同时与NB,CBA,C5.0,CMAR,CAEP,BCEP等方法的分类结果进行了比较,在UCI机器学习库中的11个数据集上的实验结果表明了文中提出的算法的有效性。Classification is one of important tasks in Data Mining. The essential Emerging Patterns are interesting knowledge patterns, can reflect some of the obvious difference between two datasets and reduce classifi- cation poison, and are suitable to solve the classification problems of dense and high dimension datasets. However, traditional classification algorithms applying sequential covering strategy failed to handle problems of small disjunction and splintered false positive. The paper proposes a improved two -phased algorithm based on eEPs, which adopts eEPs as classification patterns, builds classifier using two -phase method, and optimizes specially score strategy and the weights of two phases. Meanwhile, when compared with other traditional classification ap- proaches： NB, CBA, C5.0, CMAR, CAEP, BCEP, the effectiveness obtained by the proposed method is obvi- ously on 11 standard datasets of UCI machine learning repository.

关 键 词：数据挖掘 分类 两阶段方法 基本显露模式 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]