《计算机应用研究》|Application Research of Computers

基于BSMOTE和逆转欠抽样的不均衡数据分类算法

Classification algorithm for imbalanced data sets based on combination of BSMOTE and inverse under sampling

免费全文下载 (已被下载 次)  
获取PDF全文
作者 陈睿,张亮,杨静,胡荣贵
机构 解放军电子工程学院 网络系,合肥 230037
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2014)11-3299-05
DOI 10.3969/j.issn.1001-3695.2014.11.023
摘要 针对传统分类器在数据不均衡的情况下分类效果不理想的缺陷,为提高分类器在不均衡数据集下的分类性能,特别是少数类样本的分类能力,提出了一种基于BSMOTE和逆转欠抽样的不均衡数据分类算法。该算法使用BSMOTE进行过抽样,人工增加少数类样本的数量,然后通过优先去除样本中的冗余和噪声样本,使用逆转欠抽样方法逆转少数类样本和多数类样本的比例。通过多次进行上述抽样形成多个训练集合,使用Bagging方法集成在多个训练集合上获得的分类器来提高有效信息的利用率。实验表明,该算法较几种现有算法不仅能够提高少数类样本的分类性能,而且能够有效提高整体分类准确度。
关键词 不均衡数据集;边界少数类样本合成过抽样技术;逆转欠抽样技术;多分类器集成
基金项目 国家自然科学基金资助项目(61004069)
安徽省自然科学基金资助项目(1208085QF107)
本文URL http://www.arocmag.com/article/01-2014-11-023.html
英文标题 Classification algorithm for imbalanced data sets based on combination of BSMOTE and inverse under sampling
作者英文名 CHEN Rui, ZHANG Liang, YANG Jing, HU Rong-gui
机构英文名 Dept. of Network, PLA Electronic Engineering Institute, Hefei 230037, China
英文摘要 The result of classical classification algorithms in the case of imbalanced data sets is not satisfactory. In order to improve the classification performance under imbalanced data sets, especially the classification ability of the minority class, this paper presented a novel classification algorithm for imbalanced data sets based on combination of border synthetic minority oversampling technique (BSMOTE) and inverse under sampling. It used BSMOTE to increase the sample number of minority class, and then used a inverse under sampling method to inverse the cardinalities of the majority and minority class ratio through removing the samples of redundant and noise sample firstly. By sampling several times, it created a large number of distinct training sets. It used Bagging method to ensemble the classifiers trained on those data sets to improve the efficient use of the original data sets. Experimental results show that the proposed algorithm can not only improve classification performance in the minority class data, but also increase the overall classification accuracy rate effectively than several existing algorithms.
英文关键词 imbalanced dataset; BSMOTE; inverse under sampling; multiple classifier ensemble
参考文献 查看稿件参考文献
  [1] LI Jin-jun, WANG Can, WEI Wei, et al. Efficient mining of contrast patterns on large scale imbalanced real-life data[M] //Advances in Knowledge Discovery and Data Mining. [S. l. ] :Springer, 2013:62-73.
[2] GARCA V, SNCHEZ J S, MOLLINEDA R A. On the effectiveness of preprocessing methods when dealing with different levels of class imbalance[J] . Knowledge-Based Systems, 2012, 25(1):13-21.
[3] HE Hai-bo, GARCIA E A. Learning from imbalanced data[J] . IEEE Trans on Knowledge and Data Engineering, 2009, 21(9):1263-1284.
[4] ERTEKIN S, HUANG Jian, BOTTOU L, et al. Learning on the border:active learning in imbalanced data classification[C] //Proc of the 16th ACM Conference on Information and Knowledge Management. New York:ACM Press, 2007:127-136.
[5] CHAWLA N V, BOWYER K W, HALL L O, et al. SMOTE:synthetic minority over-sampling technique[J] . Journal of Artificial Intelligence Research, 2002, 16:321-357.
[6] ZHOU Zhi-hua, LIU Xu-ying. Training cost-sensitive neural networks with methods addressing the class imbalance problem[J] . IEEE Trans on Knowledge and Data Engineering, 2006, 18(1):63-77.
[7] LPEZ V, FERNNDEZ A, MORENO-TORRES J G, et al. Analysis of preprocessing vs. cost-sensitive learning for imbalanced classification. Open problems on intrinsic data characteristics[J] . Expert Systems with Applications, 2012, 39(7):6585-6608.
[8] ANGIULLI F. Prototype-based domain description for one-class classification[J] . IEEE Trans on Pattern Analysis and Machine Intelligence, 2012, 34(6):1131-1144.
[9] 杜娟, 姜丽丽, 陈红丽. 不均衡数据集文本分类中少数类样本生成方法研究[J] . 计算机应用研究, 2009, 26(10):3731-3734.
[10] HAN Hui, WANG Wen-yuan, MAO Bing-huan. Borderline-SMOTE:a new over-sampling method in imbalanced data sets learning[M] //Advances in Intelligent Computing. [S. l. ] :Springer, 2005:878-887.
[11] 陶新民, 张冬雪, 郝思媛, 等. 基于谱聚类欠取样的不均衡数据SVM分类算法[J] . 控制与决策, 2012, 27(12):1761-1768, 1775.
[12] CHAN P K, STOLFO S J. Toward scalable learning with non-uniform class and cost distributions:a case study in credit card fraud detection[C] //Proc of the 4th International Conference on Knowledge Discovery and Data Mining. [S. l. ] :AAAI Press, 1998:164-168.
[13] LIU Xu-ying, WU Jian-xin, ZHOU Zhi-hua. Exploratory undersampling for class-imbalance learning[J] . IEEE Trans on Systems, Man, and Cybernetics, Part B:Cybernetics, 2009, 39(2):539-550.
[14] XIE Ji-gang, QIU Zheng-ding. The effect of imbalanced data sets on LDA:a theoretical and empirical analysis[J] . Pattern Recognition, 2007, 40(2):557-562.
[15] LI Lei-jun, ZOU Bo, HU Qing-hua, et al. Dynamic classifier ensemble using classification confidence[J] . Neurocomputing, 2012, 99:581-591.
[16] 尚钢, 钟珞, 陈立耀. 神经网络结构与训练参数选取[J] . 武汉工业大学学报, 1997, 10(2):108-110.
[17] PILLAI I, FUMERA G, ROLI F. F-measure optimisation in multi-label classifiers[C] //Proc of the 21st International Conference on Pattern Recognition. [S. l. ] :IEEE Press, 2012:2424-2427.
收稿日期 2013/11/14
修回日期 2013/12/26
页码 3299-3303
中图分类号 TP301.6;TP391
文献标志码 A