Abstract: In order to solve mismatch problem of the feature selection and classifier parameters in network intrusion, this paper proposes a network intrusion detection model （F-SVM） based on coupling feature selection with classifier optimization. The evaluated standard of features is mapped into high-dimensional space by radial basis kernel function to calculate, and the relation between the network feature evaluation and network intrusion classifier is established, so the feature selection stage has solved the parameter design of the classifier, the network intrusion detection model is established and the performance is tested using KDD 99 data. The results show that F-SVM can eliminate unnecessary, redundant features, dimension of network characteristics is significantly reduced, and the optimal parameters of network intrusion classifier are obtained, which improves the network intrusion detection accuracy and detection efficiency.

Key words: features selection, classifier, network intrusion, parameter optimization, kernel function parameter

摘要： 针对网络入侵中特征选择与分类器参数不匹配问题，提出一种特征选择和分类器优化耦合的网络入侵检测模型（F-SVM）。通过径向基核函数将网络特征的评估标准映射至高维空间进行计算，建立网络特征评估和后续网络入侵分类器之间的联系，在特征选择阶段解决了分类器的参数设计问题，建立网络入侵检测模型，并采用KDD 99数据集对F-SVM的性能进行测试。结果表明，F-SVM不仅可以消除无用、冗余特征，网络特征的维数显著降低，而且获得了网络入侵分类器的最优参数，从而提高了网络入侵检测的正确率和检测效率。

关键词: 特征选择, 分类器, 网络入侵, 参数优化, 核函数参数