入侵检测中支持向量机参数选择方法

Parameter selection method of support vector machines in intrusion detection

研究了支持向量机入侵检测分类器的构造方法及惩罚因子和核参数变化对支持向量机分类性能的不同影响,提出一种递推式支持向量机参数快速选择的方法。首先选择支持向量机参数时使用支持向量率来估算推广能力,并确定最优的支持向量机参数区间;然后通过寻找特定数据空间中惩罚因子和核参数的变化规律,使用不同的规则更新惩罚因子和核参数,简化了参数选择的难度。对KDDCup99网络入侵检测数据集进行仿真实验,结果表明该方法能够快速构造支持向量机入侵检测分类器模型,同时对入侵行为具有较高的检测精度。

The effects of two parameters of the SVM,i.e.,the penalty factor and the kernel parameters,on the classification performance are studied,and the criteria of intrusion detection classifier based on SVM are designed.Firstly,we study how the support vector rate affects the variation of generality and determine the optimal section of parameters.Then we propose a new recursive method for quickly selecting the parameters of SVM,and give the steps of the SVM parameter selection.Experiments on KDD Cup 99 network intrusion detection dataset indicate that the proposed method can speed up the process of constructing an intrusion detection classifier and has a high accuracy of intrusion detection.