摘要
针对目前入侵检测系统对少数攻击类的识别率较低的问题,提出一种基于卷积神经网络和随机森林的分类方法。对CICIDS2017数据集中的数据进行预处理,并用SMOTE算法进行数据平衡;用卷积神经网络对BENIGN类和ATTACK类进行二分类,分离出ATTACK类;用PCA进行特征选择,减少特征维度,并用随机森林算法对ATTACK类进行多分类。与其他算法相比,该方法不仅增加了少数攻击类的识别率,并且对其余类的识别率也有所增加。
Aimed at the problem that the current intrusion detection system has a low recognition rate for a few attack classes,a classification method based on convolutional neural networks and random forests is proposed.The data in the CICIDS2017 data set was pre-processed,and the data was balanced using the SMOTE algorithm.The convolutional neural network was used to classify the BENIGN and ATTACK classes to separate the ATTACK classes.The PCA was used for feature selection to reduce feature dimensions,and random forest algorithm was used for multi-classification of ATTACK class.Compared with other algorithms,the proposed method not only increased the recognition rate of a few attack classes,but also increased the recognition classes of the remaining classes.
作者
张志飞
王露漫
Zhang Zhifei;Wang Luman(School of Computer and Information Technology,Beijing Jiaotong University,Beijing 100044,China)
出处
《计算机应用与软件》
北大核心
2022年第10期336-343,共8页
Computer Applications and Software
基金
国家科技支撑计划项目(2015BAG12B01-08)。
关键词
机器学习
卷积神经网络
随机森林
PCA
入侵检测
Machine learning
Convolutional neural network
Random forest
PCA
Intrusion detection
