不均衡数据分类算法的综述  

Overview of classification algorithms for unbalanced data

作　　者：陶新民[1] 郝思媛[1] 张冬雪[1] 徐鹏[1] 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《重庆邮电大学学报（自然科学版）》2013年第1期101-110,121,共11页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：国家自然科学基金(61074076);中国博士后科学基金(20090450119);中国博士点新教师基金(20092304120017)~~

摘　　要：传统的分类方法都是建立在类分布大致平衡这一假设基础上的,然而实际情况中,数据往往都是不均衡的。因此,传统分类器分类性能通常比较有限。从数据层面和算法层面对国内外分类算法做了详细而系统的概述。并通过仿真实验,比较了多种不平衡分类算法在6个不同数据集上的分类性能,发现改进的分类算法在整体性能上得到不同程度的提高,最后列出了不均衡数据分类发展还需解决的一些问题。Traditional classification methods are based on the assumption that the training sets are well-balanced,however,in real case the data is usually unbalanced,and the classification performance of the traditional classification is always restricted.A detailed overview of domestic and foreign classification algorithms from the data level and algorithm level is provided in this paper.And through simulation experiments to compare the classification performance of a variety of unbalanced classification algorithm on six different data sets,it is found that the improved classification algorithm has varying degrees of improvement for overall performance.The paper concludes with a list of problems which need solving for the development of unbalanced data classification.

关 键 词：不均衡数据 改进算法 分类性能 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]