论文部分内容阅读
提出了一种聚类学习与增量SVM训练相结合的的入侵检测方法,采用聚类分析、样本修剪与增量学习相结合的方式,通过聚合相似的训练样本以支持多类别分类,通过去除相似的样本而只取其代表点,从而减少参加训练的样本数量,提高学习效率,同时采用基于广义KKT判决的增量学习方法,有效改善了多类别入侵检测场合下样本数据集过于庞大,学习速度过慢且难以保障SVM入侵检测能力持续优化的问题。
This paper proposes a new intrusion detection method based on clustering learning and incremental SVM training. Clustering analysis, sample pruning and incremental learning are combined. By clustering similar training samples to support multi-class classification, Similar samples but only their representative points, so as to reduce the number of training samples and improve the learning efficiency. At the same time, the incremental learning method based on generalized KKT decision is used to effectively improve the sample data set under multi-category intrusion detection. Slow speed and difficult to guarantee continuous optimization of intrusion detection capability of SVM.