一种改进的SVM多类分类算法在入侵检测中的应用  

Application of an Improved SVM Multi-Class Classification to Intrusion Detection

作　　者：李太白[1] 唐万梅[1] 

机构地区：[1]重庆师范大学计算机与信息科学学院,重庆401331

出　　处：《重庆师范大学学报（自然科学版）》2012年第5期63-66,共4页Journal of Chongqing Normal University:Natural Science

基　　金：重庆市教委科学技术项目(No.KJ110617);重庆市自然科学基金(CSTC2010BB2090);重庆师范大学校级项目(No.cyjg1205)

摘　　要：入侵检测作为网络安全的关键技术,成为了当前网络安全研究的热点,入侵检测算法的准确率和推广性能是研究的重点。基于二叉树的思想和超球支持向量机的特点,本文提出了一种改进的SVM多类分类入侵检测算法。本文通过引入相似度函数作为权值,选取相似性最小的两类样本构造两类分类器,采用自下而上的方法构造多个两类超球SVM分类器,并将该多类分类算法应用于入侵检测中。利用KDD CUP 1999入侵检测数据进行了仿真实验,实验结果表明,该算法能有效提高检测准确率、推广性能也得到较好改善。Intrusion detection system as the key technology oi network security becomes research hot spot oI the current net- work security, while precision and generalization performance is the key point of intrusion detection algorithm. According to binary tree method and the characteristics of sphere structured support vector machine, an improved SVM multi-class classifi- cation algorithm is proposed to intrusion detection. This algorithm uses similarity functions as weight value and selects two kinds of sample similarity minimum to structure two-class classifier; to bottom-up structure kinds of two-class classifier of sphere structured SVM. Finally it is applied to intrusion detection. The KDD CUP 1999 intrusion detection data used to simu- late experiments. Experimental results show that the algorithm effectively improved the detection accuracy and generalization performance.

关 键 词：支持向量机 球结构 二叉树 入侵检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]