基于入侵检测的特征提取方法  

FEATURE EXTRACTION METHOD BASED ON INTRUSION DETECTION

作　　者：朱笑荣[1] 杨德运[1] 

机构地区：[1]泰山学院信息科学技术学院,山东泰安271021

出　　处：《计算机应用与软件》2010年第6期30-31,94,共3页Computer Applications and Software

基　　金：国家自然科学基金(60872161)

摘　　要：利用核主成份分析对入侵检测的训练样本进行特征提取,有效地提取出样本的分类信息,降低了维数。在此基础上,进一步将简约支持向量机RSVM(Reduced SVM)方法应用到非线性的PSVM中,降低了核矩阵的计算量。两种方法相结合提高了训练速度和入侵检测的分类效果,并且一定程度上还改善了分类的正确率和误报率,数值试验证明算法的有效性。In the paper we use the kernel principal component analysis to extract features from intrusion detection training samples.The method extracts features and reduces the dimensions very effectively.In addition,we further make use of RSVM method to nonlinear proximal SVM（PSVM） and this reduces the computation complexity of kernel matrix.The combination of the above two methods improves the training speed and the classification effect of intrusion detection.Besides,it meliorates the correction rate and false alarm rate of the classification to certain extent,numerical test also attests the validity of the algorithm.

关 键 词：RSVM PSVM 特征提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]