Abstract:
This paper presents an improved transduction network anomaly detection algorithm, it applies K-L transform for dimension reduction to high-dimensional data which is used for Euclidean distance calculation, and adopts branch and bound tree for reducing times of Euclidean distance calculation. Experiment based on KDD CUP99 dataset demonstrates improved algorithm can improve real-time performance of network anomaly detection. In comparison with anomaly detection algorithm based on OC-SVM, improved algorithm can obtain a better detection rate while keeping a proper false positive rate.
Key words:
network security,
anomaly detection,
Karhunen-Loeve(K-L) transform,
branch bound tree
摘要: 提出一种改进的直推式网络异常检测算法,利用K-L变换降低计算欧氏距离特征向量的维数,采用分支限界树剪裁减少欧氏距离的计算次数。基于KDD CUP99数据集的实验验证了改进算法能提高网络异常检测的实时性,通过与基于单类支持向量机的异常检测算法的性能对比结果表明,改进算法在保证一定误报率的情况下具有较高的检测率。
关键词:
网络安全,
异常检测,
K-L变换,
分支限界树
CLC Number:
GU Wei-Feng, WANG Yong, ZHANG Feng-Li, TONG Ban. Network Anomaly Detection Algorithm Based on Feature Compression and Branch Clipping[J]. Computer Engineering, 2010, 36(21): 137-139.
贾伟峰, 王勇, 张凤荔, 童彬. 基于特征压缩与分支剪裁的网络异常检测算法[J]. 计算机工程, 2010, 36(21): 137-139.