On trade-off between computational efficiency and prediction accuracy in bandwidth traffic estimation

The increasing demand for wireless broadband services poses the need for efficient utilisation of the backhaul network resources. To this end, schemes that use artificial neural networks in order to predict the forthcoming network traffic demand and proactively request the commitment of the necessary resources have been proposed. However, an up-to-date prediction model, required by these schemes, necessitates a regularly held training process, which incurs a high computational cost. This reported work investigates the trade-off between prediction accuracy and computational efficiency by employing evolutionary game theory and a novel scheme is proposed that can achieve both the aspects.