一种基于机器学习的MANET网络入侵检测性能评估方法研究  

Performance Analysis Method for Intrusion Detection in MANETs Based on Machine Learning Algorithms

作　　者：蒋一波[1] 王雨晨[1] 王万良[1] 张祯[1] 陈琼[1] 

机构地区：[1]浙江工业大学计算机学院,杭州310023

出　　处：《计算机科学》2013年第11A期170-174,191,共6页Computer Science

基　　金："十二五"国家科技支撑计划:农村小水电高效发电技术(2012BAD10B01)资助

摘　　要：移动Ad hoc网络(MANET,Mobile Ad hoc Networks)正得到越来越广泛的应用,相应的网络安全问题也开始得到广泛的关注。研究MANET网络可能遭遇的攻击方式,提出基于机器学习技术的入侵检测性能评估模型,并提出一个综合评价指标,比较了7种机器学习算法在MANET网络入侵检测中的性能表现,对于构建安全有效的MANET网络具有重要的意义。使用GloMoSim仿真工具对MANET网络正常行为及黑洞、洪水、丢包3种入侵行为进行模拟,并详细分析了各种攻击情况下,7种机器学习算法的性能表现。分析结果显示,该评估模型能较好地反映出各种机器学习算法的性能,其中,多层感知器、逻辑回归和支持向量机具有较高的检测率及较低的误报率。Mobile Ad-hoc network （MANET） has become an important technology in recent years and the corresponding security problems is getting more and more attention. This paper proposed a performance analysis model and an integrated evaluation index for intrusion detection based on machine learning algorithms. The experiment simulated three typical anomalous behaviors （Black hole,Flooding and Packet drop） and compared seven well-known machine learning algorithms in detail. The analysis results show that the proposed model could give a well expression to the performance of each algorithm. In particular, MultiLayer Perceptron,Logistic Regression and Support Vector Machine give the best performance and the Logistic Regression and Support Vector Machine also spend very little time to train the classification model.

关 键 词：MANET 入侵检测 机器学习 性能评估 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]