Machine Learning and Intelligent Communications. Third International Conference, MLICOM 2018, Hangzhou, China, July 6-8, 2018, Proceedings

Research Article

Smartphone Application Identification by Convolutional Neural Network

  • @INPROCEEDINGS{10.1007/978-3-030-00557-3_11,
        author={Shuang Zhao and Shuhui Chen},
        title={Smartphone Application Identification by Convolutional Neural Network},
        proceedings={Machine Learning and Intelligent Communications. Third International Conference, MLICOM 2018, Hangzhou, China, July 6-8, 2018, Proceedings},
        proceedings_a={MLICOM},
        year={2018},
        month={10},
        keywords={Application identification Convolutional neural network Mobile traffic Encrypted traffic Network management},
        doi={10.1007/978-3-030-00557-3_11}
    }
    
  • Shuang Zhao
    Shuhui Chen
    Year: 2018
    Smartphone Application Identification by Convolutional Neural Network
    MLICOM
    Springer
    DOI: 10.1007/978-3-030-00557-3_11
Shuang Zhao1,*, Shuhui Chen1,*
  • 1: National University of Defense Technology
*Contact email: Zhaos_abby@163.com, shchen@nudt.edu.cn

Abstract

Mobile traffic has received much attention within the field of network security and management due to the rapid development of mobile networks. Unlike fixed wired workstation traffic, mobile traffic is mostly carried over HTTP/HTTPS, which brings new challenges to traditional traffic identification methods. Although there have been some attempts to address this problem with side-channel traffic information and machine learning, the effectiveness of these methods majorly depends on predefined statistics features. In this paper, we presented an approach based on convolutional neural network without explicit feature extraction process. And owing to no payload inspection requirement, this method also works well even encrypted traffic appears. Six instant message applications are used to verify our approach. The evaluation shows the proposed approach can achieve more than 96% accuracy. Additionally, we also discussed how this approach performed under real-world conditions.