Traffic predictor using neural network and its performance
Hideki Tode, Atsushi Iwamoto, Hiromasa Ikeda · Electronics and Communications in Japan (Part I Communications) · 2000
In the super-high-speed network, a large amount of data must be transmitted and processed instantaneously. Thus, a technique which can examine past traffic and predict future situations is needed. This paper considers a traffic prediction technique in which a neural network is used for prediction. Several improvements for the basic neural prediction system are presented. The usefulness of the neural prediction system is demonstrated by comparison of the prediction performance to the linear prediction method, which is considered at present as an effective method of traffic prediction. More precisely, the prediction performance is evaluated in terms of the mean-square error for a video traffic model with short-term time correlation, and measured LAN traffic data for which the long-term time correlation has already been verified. For the former, neural prediction realizes almost the same prediction performance as linear prediction based on the ARMA model. For the latter, neural prediction is shown to give better performance than linear prediction by providing prediction error input that allows accurate recognition of the trend of change. © 2000 Scripta Technica, Electron Comm Jpn Pt 1, 83(11): 33–42, 2000