基于深度学习的网络入侵检测方法

A deep learning based network intrusion detection method

网络入侵检测是网络安全领域的重要课题,传统的机器学习检测算法以特征提取和特征分离为基础,存在检测能力不足和误报率高等问题。本文提出一种基于深度学习的网络入侵检测模型IDNet。其综合考虑流量数据中的空间特征和时间特征。首先使用卷积神经网络(CNN)提取流量数据的空间特征,然后通过递归神经网络(RNN)提取流量数据的时间特征,通过堆叠CNN+RNN模块,并逐步增加学习粒度,达到同时有效提取空间特征和时间特征的目的。试验结果表明,所提算法检测准确率和误报率均优于传统机器学习算法。

Network intrusion detection is a crucial subject in network security,traditional machine learning detection approaches are based on feature extraction and separation,which have defects of low detection accuracy and high false alarm rate.This paper promotes a deep learning network intrusion detection model,IDNet,which comprehensively consider the spatial and temporal features in traffi c data.Firstly,spatial features in traffi c data are extracted by convolutional neural network,then temporal features in traffi c data are extracted by recurrent neural network.The spatial and temporal features are effectively studied though a stack of CNN+RNN modules in sync with a gradually increasing granularity.Our experiments show that compared with traditional machine learning approaches,IDNet not only has a higher level of detection accuracy,but also has a much lower false alarm rate.