基于连接数据分析和OSELM分类器的网络入侵检测系统  

Network intrusion detection system based on connection data analysis and OSELM classifier

作　　者：安尼瓦尔.加马力 亚森.艾则孜 木尼拉.塔里甫 

机构地区：[1]新疆警察学院信息安全工程系,乌鲁木齐830013 [2]新疆财经大学计算机科学技术工程学院,乌鲁木齐830012

出　　处：《计算机应用研究》2017年第12期3749-3752,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61762086);国家社会科学基金资助项目(13CFX055);新疆维吾尔自治区高校科研计划重点项目(XJEDU20161052)

摘　　要：针对网络入侵的实时高效检测问题,提出一种基于网络连接数据分析和在线贯序极限学习机(OSELM)分类器的网络入侵检测系统(IDS)。对入侵数据库中的网络连接数据进行分析,通过特征选择算法选择出最优特征子集;然后,迭代执行交叉验证,并通过Alpha剖析来缩减样本尺寸,以此减低后续分类器的计算复杂度;最后,利用优化后的样本特征集来训练OSELM分类器,以此构建一个网络实时入侵检测系统。在NSL-KDD和DARPA数据库上的实验结果表明,提出的IDS具有较高的检测率和较低的误报率,同时检测时间较短,符合实时入侵检测的要求。For the problem of real-time network intrusion detection,this paper proposed a network intrusion detection system(IDS) based on network connection data analysis and online sequential extreme learning machine(OSELM) classifier. First,it analyzed the network connection data in the intrusion database and selected the optimal feature subset by the feature selection algorithm. Then,it executed a iterative implementation of cross-validation,and through the Alpha profile to reduce the sample size so as to reduce the complexity of the subsequent classifier calculation. Finally,it trained the OSELM classifier by using the optimized sample feature set to construct a real-time network intrusion detection system. Experimental results on the NSL-KDD and DARPA database show that the proposed IDS has a high detection rate and a low false alarm rate,and the detection time is short,which meets the requirements of real-time intrusion detection.

关 键 词：入侵检测系统 网络连接数据 特征选择 在线贯序极限学习机 Alpha剖析 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术]