Multi-scale high-speed network traffic prediction using combination of neural networks

High-speed network traffic prediction is considered as the core of the preventive congestion control. In this paper, we apply two different artificial neural network (ANN) architectures, multilayer perceptron (MLP) and fuzzy neural network (FNN), to predict one-step ahead the value of the MPEG and JPEG video, Ethernet and Internet traffic data. To enhance prediction accuracy, the output of the individual ANN predictors are combined using different combination schemes. An adaptive updating scheme is used in both of the ANNs and combination schemes. This adaptation makes the predictors dynamic and allows them to capture the non-stationary traffic characteristics. Prediction at different time scales is considered in order to apply the predicted values to the congestion control schemes. The results show that the ANN predictors outperform the autoregressive (AR) model, and the combination approach enhances the prediction accuracy.