SW-Filtering: An Efficient Traffic Classification Method Based on Convolutional Neural Networks
Abstract— We propose a compact and effective CNN(convolutional neural network)-based method for real time traffic classification. Firstly, we apply a stochastic strategy in the convolutional pooling filter, so as to prevent overfitting and enhance local features. We implement the CNN architecture with one-dimensional convolutional kernel. Furthermore, we stochastically select flows from the BoFs (bag of flows) to classify in each time period and record each classification result. Finally, by taking all these results into voting, we obtain the final results for each BoF (all the flows in one BoF share one label). This method not only prevents some inherent problems of CNN-based classifiers, such as overfitting and patterns down-weighting, but is also efficient in memory consumption. Our mathematical proofs and experiments demonstrated that the proposed method can significantly improve the performance of CNN-based traffic classifiers in both accuracy and space efficiency.
Index Terms— Deep learning; traffic classification; convolutional neural networks; stochastic window filtering;
He Huang, Haojiang Deng, Jun Chen, Zhichuan Guo, Runzi Zhang
National Network New Media Engineering Research Center, Institute of Acoustics, University of Chinese Academy of Science, CHINA
Cite: He Huang, Haojiang Deng, Jun Chen, Zhichuan Guo, Runzi Zhang, "SW-Filtering: An Efficient Traffic Classification Method Based on Convolutional Neural Networks," Proceedings of 2018 the 8th International Workshop on Computer Science and Engineering, pp. 288-292, Bangkok, 28-30 June, 2018.