Premium
Deep learning–based network application classification for SDN
Author(s) -
Zhang Chuangchuang,
Wang Xingwei,
Li Fuliang,
He Qiang,
Huang Min
Publication year - 2018
Publication title -
transactions on emerging telecommunications technologies
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.366
H-Index - 47
ISSN - 2161-3915
DOI - 10.1002/ett.3302
Subject(s) - softmax function , computer science , artificial intelligence , traffic classification , autoencoder , machine learning , feature selection , classifier (uml) , artificial neural network , deep learning , data mining , feature extraction , support vector machine , pattern recognition (psychology) , the internet , world wide web
Accurate application classification is important and useful to improve network performance. However, with the continuous expansion of network scale and the rapid increase of network users, it is very difficult for the existing application classification methods to accurately identify and classify network applications. Currently, most classification methods are suitable for small‐scale data sets and cannot achieve high classification accuracy because of the shallow learning structure and the limited learning ability. The emergence of deep learning technology and software‐defined networking (SDN) enables the application classification method to process large‐scale data. In this paper, by leveraging the SDN architecture, we present a novel hybrid deep neural network–based application classification method, which achieves high classification accuracy without the manual feature selection and extraction. In the proposed application classification framework, by taking the advantage of the logical centralized control and powerful computing capability, the massive network traffic is easily collected and processed by the SDN controller. The processed data is used to train the hybrid deep neural network, which is composed of the stacked autoencoder and softmax regression layer. The deep flow features can be obtained from the stacked autoencoder automatically instead of the manual feature selection and extraction. The softmax regression layer is used as the classifier to realize the application classification. Finally, simulation results demonstrate that our proposed classification method is effective and gets higher classification accuracy than the support vector machine–based classification method.