适用于云环境的LASVM异常检测研究

Research on Anomaly Detection of Lasvm in Cloud Environment

在云环境下,数量众多的虚拟机导致的海量数据和实时检测的需求使得传统SVM不再适合作为异常检测算法。文中提出使用基于SMO算法改进的具有在线学习能力的LASVM算法进行异常检测。在使用过抽样平衡后的KDD CUP99数据集和Lib SVM的对比试验中,证明了该算法在接近Lib SVM精确度的同时对内存需求量更小,并且训练速度对缓存大小不敏感,能够处理海量数据,且该算法具有在线学习能力,适合在云环境下做实时异常检测的分类器。

In the cloud environment,a large number of virtual machines lead to massive data,as well as real-time detection needs,making traditional SVM no longer suitable for anomaly detection algorithms. In this paper,an improved LASVM algorithm with online learning ability based on SMO algorithm is proposed for anomaly detection. In the contrast test using balanced KDD CUP99 data set after over-samplingwith Lib SVM,experimental evidence indicates that LASVM is near Lib SVM accuracy while the memory demand is smaller,and the training speed is not sensitive to the size of the cache,is able to deal with huge amounts of data,and the algorithm has online learning ability,suitable for Real-time Anomaly Detection classifier in the cloud environment.