Deep learning approach to multimedia traffic classification based on QoS characteristics. Issue 3 (1st May 2019)

• Record Type: Journal Article
• Authors: Wang, Zaijian; Mao, Shiwen; Yang, Weidong
• Is Part Of: IET networks. Volume 8:Issue 3(2019)
• Page Start: 145
• Page End: 154
• Publication Date: 2019-05-01
• Subjects: quality of service -- learning (artificial intelligence) -- pattern classification -- Internet of Things -- statistical analysis -- multimedia communication -- telecommunication traffic; training data conditions -- shallow classification methods -- emerging multimedia traffic types -- big data -- multimedia traffic classification problem -- deep learning approach -- improved DL‐based multimedia traffic classification method -- multimedia data -- relevant QoS features -- multimedia datasets -- QoS characteristics -- Things applications -- IoT traffic -- traffic flow data -- QoS perspective; Computer network architectures -- Periodicals; Computer network protocols -- Periodicals; Information networks -- Periodicals; Telecommunication systems -- Periodicals; 004.605
• DOI: 10.1049/iet-net.2018.5179
• Languages: English
• ISSNs: 2047-4954

Abstract

Abstract : With the fast increase of multimedia traffic in Internet of Things (IoT) applications, IoT traffic now requires very different Quality of Service (QoS). By extensive statistical analysis of traffic flow data from a real world network, the authors find that there are some latent features hidden in the multimedia data, which can be useful for accurately differentiating multimedia traffic flows from the QoS perspective. Under limited training data conditions, existing shallow classification methods are limited in performance, and are thus not effective in classifying emerging multimedia traffic types, which have truly entered the era of big data and become very completed in QoS features. This situation inspires us to revisit the multimedia traffic classification problem with a deep learning (DL) approach. In this study, an improved DL‐based multimedia traffic classification method is proposed, which considers the inherent structure of QoS features in multimedia data. An improved stacked autoencoder model is employed to learn the relevant QoS features of multimedia traffic. Extensive experimental studies with multimedia datasets captured from a campus network demonstrate the effectiveness of the proposed method over six benchmark schemes.