基于MRMHC-LSVM的IP流分类  

IP flow classification based on MRMHC-LSVM

作　　者：李文法[1,2] 段洣毅[1,2] 陈友[1,2] 程学旗[1,2] 

机构地区：[1]中国科学院计算技术研究所,北京100190 [2]中国科学院研究生院,北京100039

出　　处：《高技术通讯》2009年第6期564-571,共8页Chinese High Technology Letters

基　　金：863计划(2006AA01Z452,2007AA01Z416);国家242信息安全计划(2005C39)资助项目

摘　　要：提出了一种构建轻量级的IP流分类器的wrapper型特征选择算法MRMHCLSVM。该算法采用改进的随机变异爬山(MRMHC)搜索策略对特征子集空间进行随机搜索,然后利用提供的数据在无约束优化线性支持向量机(LSVM)上的分类错误率作为特征子集的评价标准来获取最优特征子集。在IP流数据集上进行了大量的实验,实验结果表明基于MRMHC-LSVM的流分类器在不影响分类准确度的情况下能够提高检测速度,与当前典型的流分类器NBK-FCBF相比,基于MRMHC-LSVM的IP流分类器具有更小的计算复杂度与更高的检测率。This paper proposes a wrapper feature selection algorithm MRMHC-LSVM aiming at modeling lightweight flow classi- tiers by using the modified random mutation hill climbing （MRMHC） approach as the search strategy to specify candidate subsets for evaluation and using the linear support vector machine （LSVM） algorithm as the wrapper approach to obtain the optimum feature subset. The feasibility of the algorithm was examined by conducting several experiments on flow datasets. The experimental results show that a classifier based on the proposed approach can greatly improve the computa- tional performance without negative impact on the classification accuracy. Further more, this approach is able not only to have smaller resource consumption, but also to have higher classification accuracy than the Naive Bayes method with Kernel density estimation after Fast Correlation-Based Filter （NBK-FCBF）.

关 键 词：流分类 特征选择 改进的随机变异爬山(MRMHC) 线性支持向量机(LSvM) 

分 类 号：TN915.01[电子电信—通信与信息系统]