基于改进深度卷积神经网络的网络流量分类方法

Network traffic classification method based on improved deep convolutional neural network

机器学习方法对网络流量分类的前提是假设流量具有独立同分布性,而实际情况下流量特征不断发生变化,导致该方法在处理海量、不具备独立同分布的流量数据时开销较大,计算复杂度较高,精度较低.针对上述问题,本文提出一种新的分类模型.该模型将PCA算法与改进的深度卷积神经网络分类模型(improved deep LeNet-5 convolutional neural networks,LCNN)相结合进行流量分类.前者进行降维分析,发现影响检测精度的关键特征,后者采用自主特征学习方式提升分类精度.实验表明,本文方法的内存开销较之前方法降低了3.2%,检测精度提升了5%~8%.

The prerequisite of traffic classification based on machine learning models is that traffic is independent and has identical distributions.However,traffic changes in the wild increase the memory cost of these models and reduces their accuracy.To tackle these problems,this work proposes a new classification model.The model combines a principal component analysis algorithm and an improved deep convolutional neural network.The former performs dimensionality reduction so that the key features affecting detection accuracy are found.The latter adopts the autonomous feature learning method to improve the classification accuracy.Experiments show that the memory overhead is reduced by 3.2%and that the detection accuracy is improved by 8%relative to other similar works.