摘要
支持向量机所具有的处理小样本和良好的推广能力的优势,在入侵检测中得到了广泛应用。考虑到数据特征的高维性和冗余性,特征提取是一个关键步骤。采用非线性流形学习算法L-Isomap对入侵检测数据进行特征选择,然后应用one-classSVM训练并识别异常。通过将异构值差度量(HVDM)距离代替欧几里德距离提出了HL-Isomap。选用KDD数据集来比较上述不同模型,实验结果表明了降维方法的有效性,尤其是误警率性能得到了显著的提高。
With great advantages in small sample and machine generalization ability,support vector machine has been widely applied in intrusion detection.Due to high dimensionality and redundancy of data,feature extraction is a crucial procedure.This paper proposes a scheme using popular non-linear dimension reduction tool L-Isomap and one-class support vector machine to detect intrusions.HL-Isomap is also proposed through replacing Euclidean metric with heterogeneous value difference metric.This paper evaluates different models with the KDD dataset.The experiment results show that the dimension reduction method is effective and the proposed model outperforms the conventional one-class SVM in false positive rate.
出处
《计算机工程与应用》
CSCD
北大核心
2010年第28期85-87,共3页
Computer Engineering and Applications
基金
教育部科学技术研究重点(重大)项目No.107021~~
