一种高效的的网络入侵检测特征提取方法

在网络入侵检测中，样本数据的特征维数较高，而冗余特征的存在使系统的存储负担加重，分类器性能降低。本文提出一种基于Fisher Score和SVM的特征重要性度量和提取方法，针对KDD'99网络入侵检测数据集，应用该方法得到了混合攻击和单一攻击模式下的特征重要度排序，选取重要特征建立SVM入侵检测分类器，结果表明分类器精度与使用全部特征构建的SVM分类器相当，训练和测试时间有显著降低。

An efficient Network Intrusion Detection Feature Extraction Method

In network intrusion detection system, the feature dimension is high. But redundant features will increase the cost of storage and decrease the performance of classifier. Fisher score and SVM Based network intrusion detection feature extraction algorithm was presented in this paper. In accordance with KDD'99 standard intrusion detection dataset, the feature significance ranking for mixed attack and four single attacks were obtained by using this method. By extracting the important features to build SVM classifier, experiment results show, the accuracy of the classifier is approximate equivalent with the classifier constructed by all features, the training and testing time decrease dramatically.