基于特征优化和BP神经网络的入侵检测方法  

Intrusion detection method based on feature optimization and BP neural network

作　　者：王伟[1] 代红[1] 赵斯祺 WANG Wei;DAI Hong;ZHAO Si-qi(School of Computer Science and Software Engineering,University of Science and Technology Liaoning,Anshan 114051,China)

机构地区：[1]辽宁科技大学计算机与软件工程学院,辽宁鞍山114051

出　　处：《计算机工程与设计》2021年第10期2755-2761,共7页Computer Engineering and Design

基　　金：2019年辽宁省科学技术基金项目(自然基金指导计划)(2019-ZD-0270)。

摘　　要：为提高网络入侵检测率,提出一个集特征优化和人工神经网络于一体的网络入侵识别发现框架AS-BP。引入SMOTE技术和随机采样技术对数据进行平衡约简处理,解决数据不平衡问题,利用集成方法对网络入侵数据进行重要特征提取,降低数据处理维度,通过优化BP神经网络算法,对网络入侵数据进行判断完成分类。实验结果表明,该方法克服了传统BP神经网络建模时间过长的问题,在不降低其它攻击类型检测率的同时,提高U2R和R2L的检测率,克服了数据集中少数类数据量过少导致的少数类检测率低的问题。将实验结果与其它分类方法进行比较,验证了该方法的准确率、精确率和召回率优于其它方法。To improve the detection rate of internet intrusion detection,a network intrusion recognition and discovery framework AS-BP was proposed.Feature optimization and artificial neural network were integrated.To solve the problem of unbalance data,SMOTE technology and random sampling technology were introduced.They were used to balance and reduce data.Important features of network intrusion data were extracted from integration method,which were used to reduce the dimension of data processing.A classification test experiment was performed on the network intrusion data using the optimized BP neural network.Experimental results show that the proposed method overcomes the problem of long modeling time,and improves the detection rate of U2R and R2L without reducing the detection rate of other attack types.The improved algorithm overcomes the problem of low detection rate of the minority class caused by the small amount of the minority class in the dataset.The results of experiment verify that the classifier is better than several other classifiers in accuracy,precision and recall.

关 键 词：入侵检测 SMOTE技术 随机采样技术 集成方法 BP神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]