Network Traffic Prediction Using Recurrent Neural Networks

Abstract

The network traffic prediction problem involves predicting characteristics of future network traffic from observations of past traffic. Network traffic prediction has a variety of applications including network monitoring, resource management, and threat detection. In this paper, we propose several Recurrent Neural Network (RNN) architectures (the standard RNN, Long Short Term Memory (LSTM) networks, and Gated Recurrent Units (GRU)) to solve the network traffic prediction problem. We analyze the performance of these models on three important problems in network traffic prediction: volume prediction, packet protocol prediction, and packet distribution prediction. We achieve state of the art results on the volume prediction problem on public datasets such as the GEANT and Abilene networks. We also believe this is the first work in the domain of protocol prediction and packet distribution prediction using RNN architectures. In this paper, we show that RNN architectures demonstrate promising results in all three of these domains in network traffic prediction, outperforming standard statistical forecasting models significantly.