A Neural Network Approach for Predicting Network Resource Requirements in Video Transmission Systems

Dynamic resource allocation is important for ensuring ef-ficient network utilization in Internet-based multimedia
content delivery system. To allow accurate network traf-
fic prediction in the case of video delivery, relevant infor-
mation based on video contents and the short term traffic
pattern has to be taken into account, while the inclusion of
non-relevant features will deterioriate the prediction per-
formance due to the "curse of dimensionality" problem.
In this work, we propose a neural network-based predic-
tion system and specifically address the determination of
relevant input features for the system. Experiments have
shown that the current system is capable of identifying
a highly relevant subset of features for traffic prediction
given a large number of video content and short-term net-
work traffic descriptors.