计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (3): 94-99.DOI: 10.3778/j.issn.1002-8331.1905-0139

• 网络、通信与安全 • 上一篇    下一篇

基于一维卷积神经网络的网络流量分类方法

李道全,王雪,于波,黄泰铭   

  1. 青岛理工大学 信息与控制工程学院,山东 青岛 266520
  • 出版日期:2020-02-01 发布日期:2020-01-20

Network Traffic Classification Method Based on One-Dimensional Convolution Neural Network

LI Daoquan, WANG Xue, YU Bo, HUANG Taiming   

  1. School of Information and Control Engineering, Qingdao University of Technology, Qingdao, Shandong 266520, China
  • Online:2020-02-01 Published:2020-01-20

摘要: 针对传统机器学习算法对于流量分类的瓶颈问题,提出基于一维卷积神经网络模型的应用程序流量分类算法。将网络流量数据集进行数据预处理,去除无关数据字段,并使数据满足卷积神经网络的输入特性。设计了一种新的一维卷积神经网络模型,从网络结构、超参数空间以及参数优化方面入手构造了最优分类模型。该模型通过卷积层自主学习数据特征,解决了传统基于机器学习的流量分类算法中特征选择问题。通过网络公开数据集进行模型测试,相比于传统的一维卷积神经网络模型,所设计的神经网络模型的分类准确率提升了16.4%,总分类时间节省了71.48%。另外在类精度、召回率以及[F1]分数方面都有较好的提升。

关键词: 一维卷积神经网络, 流量分类, 数据预处理, 参数优化, 深度学习

Abstract: Aiming at the bottleneck problem of traditional machine learning algorithm for traffic classification, an application traffic classification algorithm based on one-dimensional convolution neural network model of deep learning algorithm is proposed. The data set of network traffic is preprocessed. In the data preprocessing stage, the irrelevant data fields are removed and the input characteristics of convolution neural network are satisfied. A new one-dimensional convolution neural network model is proposed, and the optimal classification model is constructed from the aspects of network structure, hyper-parametric space and parameter optimization. This model solves the problem of feature selection in traditional traffic classification algorithm based on machine learning by self-learning data features in convolution layer. Compared with the traditional one-dimensional convolution neural network model, the designed neural network model improves the classification accuracy by 16.4%, and the total classification time is saved by 71.48%. In addition, the class accuracy, recall rate and F1 score have been improved.

Key words: one-dimensional convolution neural network, network traffic classification, data preprocessing, parameter optimization, deep learning