摘要
为了更好地改进集成分类器的性能,提出了一种基于反馈学习的支持向量机Bagging集成分类算法.该算法在对子分类器的训练中,引入反馈学习的思想,首先对每个训练得到的子分类器进行测试,找到被错分的样本,把这些样本添加到训练样本集中,重新进行训练、测试,直到没有新的被错分的样本出现为止,最后采用多数投票策略对得到的各子分类器进行组合.仿真实验结果表明,该算法可通过提高各分类器的分类能力改进集成学习器的性能.
Performance of an ensemble SVM depends on its base SVMs and their combinational methods. To improve the ensemble SVM,a novel algorithm by using SVMs Bagging based on feedback learning are presented in this paper. This algorithm introduces the thought of feedback learning into the training of the sub-classifier. Firstly,a sub-classifier is trained with a training subset that is drawn randomly from the original training set, and then find misclassified examples according to validation. Secondly, these misclassification examples are added to the training subset and retrained until no one of new misclassification examples is found once again. Finally,the majority-vote scheme is used to combine all the learned support vector machine classifiers. The imitation experiment results on UCI database showed the better performance of the method.
出处
《甘肃农业大学学报》
CAS
CSCD
2008年第1期147-150,共4页
Journal of Gansu Agricultural University
