摘要
针对支持向量机理论中存在的问题:训练样本数量多以及必须满足Mercer条件等,提出了一种基于相关向量机(RVM)的网络入侵检测方法。首先采用"删除特征"法对KDD 99数据集中的41个特征进行评级,筛选出针对不同入侵类型的重要特征和非重要特征,然后只选择重要特征进行匹配。结果表明,这种方法与基于支持向量机(SVM)的入侵检测模型相比,具有更高的检测率和更低的误警率。
For some problems in support vector machine theory, such as the large amount of training samples and the necessity to satisfy the Mercer conditions, etc. a new method based on relevant vector machine algorithm for network intrusion detection is proposed. First, the ＂feature deduction＂ method is applied to rating the 41 features in the KDD 99 dataset, and the important features and unimportant features are selected according to different attack types, and only the important features could, in IDS, effectively increase the detection rate and reduce the false alarm rate and the detecting time. Comparison with the SVM-based model indicates that the proposed RVM-based model is of higher detection probability and much lower false alarm rate.
出处
《信息安全与通信保密》
2010年第8期47-48,51,共3页
Information Security and Communications Privacy
基金
国家自然科学基金资助项目(批准号:60802058)
教育部留学回国人员科研启动基金资助项目
关键词
入侵检测
支持向量机
相关向量机
KDD
99数据集
intrusion detection
support vector machine
relevance vector machine
KDD 99 dataset
