Evaluation of network traffic prediction based on neural networks with multi-task learning and multiresolution decomposition

TLDR: Comparisons of predictions produced by different types of neural networks with forecasts from statistical time series models show that nonlinear prediction based on NNs is better suited for traffic prediction purposes than linear forecasting models.

Abstract: Network traffic exhibits strong correlations which make it suitable for prediction. Real-time forecasting of network traffic load accurately and in a computationally efficient manner is the key element of proactive network management and congestion control. This paper compares predictions produced by different types of neural networks (NN) with forecasts from statistical time series models (ARMA, ARAR, HW). The novelty of our approach is to predict aggregated Ethernet traffic with NNs employing multiresolution learning (MRL) which is based on wavelet decomposition. In addition, we introduce a new NN training paradigm, namely the combination of multi-task learning with MRL. The experimental results show that nonlinear prediction based on NNs is better suited for traffic prediction purposes than linear forecasting models. Moreover, MRL helps to exploit the correlation structures at lower resolutions of the traffic trace and improves the generalization capability of NNs.