Network Traffic Classification using Deep Neural Networks

Network traffic control engineers are actively researching the problem of analyzing traffic of network to detect protocols, services, and applications. It is an inevitable task for ISPs and Telcos to keep an eye on what kind of traffic protocols are running on their networks to formulate their quality of service (QOS) and marketing strategies. Earlier research has found machine learning techniques that could facilitate the identification of applications and services. In this research, predominantly a dataset has been created, which is collected on the university network at various times and designed to have as much of network flows and protocols running on the university network. Dataset is then pre-processed to extract features values, remove any incomplete flows and for SNI based labeling. We used three high level features (Inter-arrival time, Payload and Packet size) sequences and makes use of a group of deep learning architectures based on CNN and RNN. We also compared our model with Random Forest Classifier. We trained our model on our own dataset and results show a high-end accuracy of 99%. We have compared both Random Forest model with our deep learning (DL) model and results show that DL model outperforms the traditional ML model.