The Effect of Using Attribute Information in Network Traffic Prediction with Deep Learning

Abstract

It is crucial for network operators to predict network traffic in the future as accurate as possible for appropriate resource provisioning and traffic engineering. Recurrent neural network (RNN) methods are considered to be the most promising prediction methods because of their high prediction accuracy. In conventional studies, RNN methods use only time series of traffic volume as input, and do not use any attribute information (e.g., timestamp and day of the week) of the time series data. However, traffic volume changes depending on both time and day of the week. Therefore, it is possible that we can further improve the prediction accuracy of the RNN methods by using the attribute information as input, in addition to the time series of traffic volume. In this paper, we investigate the effect of using the attribute information of time series of traffic volume on prediction accuracy in network traffic prediction. We propose two RNN methods: RNN-VT method and RNN-VTD method. The RNN-VT method uses timestamp information and the RNN-VTD method uses both timestamp and day of the week information as input, in addition to the time series of traffic volume. Experimental results show that day of the week information is significantly effective for improving prediction accuracy of the RNN methods while timestamp information is not effective.