PhD Forum: Data traffic classification using deep learning models

Abstract

The growth in data traffic is exponential with the usage of network applications in mundane activities. The network operators are posed with the challenge of providing Quality of Experience (QoE) to the deluge of internet users. Network traffic classification plays a significant role in network resource management with prominent security, billing, and accounting applications. In this paper, the network data traffic classification is performed using the Deep Learning (DL) models. The previous data traffic mechanism fails when the scale in data generated is in Exabytes per day from the perspective of the internet user. The network resource management is automated by classification of the traffic without intervening by the operator. The dataset is collected from the campus network for different applications.The network traffic classification is performed using the AlexNet, ResNet, and GoogLeNet DL models. The accuracy obtained for ResNet is 95%, AlexNet is 75%, and GoogLeNet is 91%. The challenge in network traffic classification is converting the packet capture files to the data type requirements of the Convolution Neural Networks (CNN) as an image. The four different network applications are considered for traffic classification. In the next-generation network architecture, artificial intelligence will be an integral part. It reduces the human intervention in traffic characterization and analysis, which aids in meeting the QoE requirements of the users.