【作者】 施贝； 杨榆；

【Author】 Shi Bei;Yang Yu;Information Security Centerof Beijing University of Posts and Telecommunications;

【机构】 北京邮电大学信息安全中心；

【摘要】 现有大部分的异常检测系统都是把数据分成正常和异常两类,这样可能会丢失重要信息,也有一些对于多分类的研究,但效果不太理想。另外网络入侵检测数据集中存在的大量冗余和噪声特征会影响检测系统的性能。针对上述两个问题,本文提出了一种基于Fisher分和多分类支持向量机的入侵检测特征选择算法。该方法先通过研究四个二分类并根据各特征的Fisher分值大小排序,得到四个特征降序序列。再根据这些序列结合多分类支持向量机分类算法,建立特征分类模型,筛选出一个最优特征组合。仿真测试结果表明,该方法具有较高的检测率和较低的测试时间,提高了系统性能。更多还原

【Abstract】 The most Intrusion detection systems divided data into two classes,which are normal and abnormal,so that it might lose some important infoimation,even though there are random researches based on multi- class,but those results are not so good.In addition there are many redundant and noisy characteristics in network intrusion detection data set,which leads to a bad performance of the detection system.To solve the above two problems,an intrusion detection feature selection algorithm based on fisher value and multi- class support vector machines was proposed.Firstly the method sorted each feature in descending order by its fisher value through researching four two classification,in return four descending sequences of feature were obtained.Then by combining the multi- class support vector machines and the sequences,and establishing classification model,the optimal feature subset was selected.The simulation test results show that the method can improve the detection accuracy,reduce the testing time,and improve the performance of the systems.更多还原