【作者】 张红梅； 高海华； 王行愚；

【Author】 ZHANG Hong-mei1,2,GAO Hai-hua1,WANG Xing-yu1(1.School of Information Science and Engineering,East China University of Science & Technology,Shanghai 200237,China;2.School of Information and Communication,Guilin University of Electronic Technology,Guilin 541004,China)

【机构】 华东理工大学信息科学与工程学院；

【摘要】 AdaBoost通过对错分样本增加权重来调整在后续单分类器序列中这些样本的重要度,使得错分样本能被正确分类,从而提升学习精度。但如果样本点存在噪声或错误,会导致最终的分类器集成缺乏稳定性,泛化能力下降。针对这个问题,提出了一种权值阈值设定的方法来限制噪声样本的权值上限,并用入侵检测数据对算法改进前后进行了评估。实验结果表明,噪声样本抑制的AdaBoost算法具有很强的稳定性和泛化能力。更多还原

【Abstract】 The intuitive idea of AdaBoost is that misclassified samples get higher weights in the next iterations,which drives the classifiers of later iteration focus on the these samples,so the learning accuracy is boosted.However,if the noise levels of samples are high,AdaBoost will become unstable and the generlization ability will decrease.To solve the problem,a noise resticted AdaBoost algorithm is proposed to bound the weights of noisy samples,and then the intrusion detection datasets are used to compare the performance of AdaBoost algorithm with NS-Adaboost.The results show that the NS-AdaBoost algorithm bears high stablility and strong generlization ability.更多还原