南京大学学报(自然科学版) ›› 2013, Vol. 49 ›› Issue (2): 150158.
刘小平1,徐桂云1,任世锦2**,杨茂云1·2
Liu Xiao一Ping1,Xu Gui-Yun1 ,Ren Shi一Jin2 ,Yanh Mao-Yun1.2
摘要: 分析了多类支持向量数据描述(support vector data description,SVDD)算法存在的问题,提出
一种新的不平衡数据二一NSVDD多分类算法.该方法借鉴了二SVM方法以及带有负类的SVDD的思
想,并基于不同类别样木间隔最大原理,较好地克服噪声和在野点的影响,提高了分类模型的泛化性能;
通过样木加权的方法解决了不平衡类别样木预测精度低的问题,并在理论上给出了根据类别样木数量
设置样木加权系数的方法.针对实际应用存在大量复杂、非线性分类数据,通过核方法把上述线性分类
算法推]’一到非线性数据分类情形.由于现有的多分类器无法实现拒判,而且每个分类器的核函数参数不
同,导致数据点与各个超球中心距离的计算结果与实际距离不相符,影响了数据判决结果的准确性和可
靠性.针对上述问题,给出基于相对距离和K-NN规则相结合的多分类方法,提高了分类结果的准确性
和可靠性.使用Benchmark数据集进行仿真实验,结果表明木算法能够获得较低的分类误差,能够有效
处理样木不平衡问题.
[1]Huang G X,Chen H F, Zhou Z L, et al. Two- class support vector data description. Pattern Recognition, 2011,44:320一329. [2]Zhu M L,Liu X D,Chen S F. Solving the problem of multi class pattern recognition with sphere- structured support vector machines. Journal of Nanjing University(Natural Sciences),2003,39 (2) ; 153-158.(朱美琳,刘向东,陈世福.用球结构的支持向量机解决多分类问题.南京大学学报 (自然科学),2003,39(2);153-158). [3]Issam B K,Claus W, Mohamcd L. Kernel k- means clustering based local support vector do- main description fault detection of multimodal processes. Expert Systems with Applications,2012,39:2166一2171. [4]Sakla W , Chan A,Ji J,et al. An SVDD-based al gorithm for target detection in hyperspectral im agery. IEEE Transactions on Geoscience and Re mote Sensing Letters,2011,8(2):384一388. [5]David M J T,Duin R P W. Support vector data description. Machine Learning,2004,54.45一66. [6]Guo S M, Chen L C,Tsai J S H. A boundary method for outlier detection based on support vector domain description. Pattern Recognition, 2009,42:77一83. [7]Zhu X K,Yang D G. A multiclass support vector domain description for pattern recognition based on a measure of expansibility. Acta Electronica Sinica,2009,37(3):464-469.(朱孝开,杨德贵. 基于推广能力测度的多类SVDD模式识别方法. 电子学报, 2009,37(3):464-469). [8]Mu T T,Nandi A K. Multiclass classification based on extended support vector data description, IEEE Transactions on System, Man, and Cybernetics一Part B;Cybernetics,2009,39(5):1206一1212. [9]Daewon L,Jaewook L. Domain described support vector classifier for multi-classification problems. Pattern Recognition, 2007 , 40 ; 41一51. [10]Tao Q, Wu G, Wang J. A new maximum maegin algorithm for one-class problems and its boosting implemention. Pattern Recognition, 2005,38 (10):1071一1077. [11]Wei X K,Lofberg J,Feng Y,et al. Enclosing ma- chine learning for class description. Lecture Notes in Computer Science. Springer-Verlag, 2007,4491:424一433. [12]Dolia A N, Harris C J,Shawe T J,et al. Kernel ellipsoidal trimming. Computational Statistic and Data Analysis,2007,52(1):309一324. [13]Feng A M, Xue H,Liu X J,et al. Enhanced one class support vector machine. Journal of Comput- er and Development,2008,45(11):1858一1864. (冯爱民,薛晖,刘学军等.增强型单类支持向 量机.计算机研究与发展,2008,45(11);1858 - 1864). [14]Lee K Y,Kim D W, Lee K H,et al. Dcnsity-in-duced support vector data description, IEEE Transactions on Neural Networks, 2007,18(1):284~289. [15]Wang W L,Wang Z Y,Zheng J W,et al. Dcnsity-in- duced data description one-class classficr. Control and Decision, 2001 , 26 ( 11) ; 1665 -1669.(王万良,王震宇,郑建炜等.密度诱导型数据描述单类分类机.控制与决策,2011,26(11);1665-1669). [16]Zhao F,Zhang J Y,Liu J. An optimization kernel algorithm for improving the performance of sup- port vector domain description. Acta Automatica Sinca, 2008 , 34 ( 9 ) : 1122 - 1127.(赵峰,张军 英,刘敬.一种改善支撑向量域描述性能的核优化算法.自动化学报,2008, 34( 9 ) : 1122 - 1127). [17]Liao Z M, Hu G Y,Zhao L W,et al. Support vec for data description implemented in class-imbal- ance learning. Journal of Applied Science- Electronics and information Engineering,2008,26 (1):79-84. (缪志敏,胡谷雨,丁力等.SVDD 在类别不平衡学习中的应用.应用科学学报,2008,26(1):79一84). [18]Michied J S,Taylor C C. Machine learning; Neu- ral and statistical classification, http;//www. ncc. up. pt/iacc/MI./sta tlog/data. html,2011- 03-12. [19]Zheng F H,Xu H,Li P,et al. Mining knowlwdge from unblanced data based on v-Support vector machine. Journal of Zhejiang University(Fngi- neering Science),2006,40(10):1682一1887.(郑 恩辉,许宏,李平等.基于v-SVM的不平衡数据挖掘研究.浙江大学学报(工学版),2006,40 (10):1682一1889). |
No related articles found! |
|