一种基于凸壳算法的SVM集成方法  

SVM Ensemble Approach Based on Convex-hull Algorithm

作　　者：张宏达[1] 王晓丹[1] 白冬婴[1] 刘倞源[1] 

机构地区：[1]空军工程大学导弹学院,三原713800

出　　处：《计算机工程》2008年第17期28-30,共3页Computer Engineering

基　　金：国家自然科学基金资助项目(50505051);陕西省自然科学基金资助项目(2007F19);空军工程大学导弹学院研究生学位论文创新基金资助项目(DY06102)

摘　　要：为提高支持向量机(SVM)集成的训练速度,提出一种基于凸壳算法的SVM集成方法,得到训练集各类数据的壳向量,将其作为基分类器的训练集,并采用Bagging策略集成各个SVM。在训练过程中,通过抛弃性能较差的基分类器,进一步提高集成分类精度。将该方法用于3组数据,实验结果表明,SVM集成的训练和分类速度平均分别提高了266%和25%。To improve the training speed of Support Vector Machine（SVM） ensemble, this paper proposes a new approach of SVM ensemble using convex-hull algorithm. The approach applies convex-hull algorithm to get from each class the hull vectors and takes these hull vectors as the training dataset for every base-classifier, Bagging method is used to aggregate the base-classifiers. Threshold is set to discard the base-classifiers with weak performance in training the ensemble to further improve the classification accuracy. Experimental results obtained from applying the proposed approach to 3 different datasets indicate that on average it accelerates training by 266% and speeds up classifying by 25%.

关 键 词：凸壳算法 支持向量机 集成 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]