A wavelet-based method to predict Internet traffic

Abstract

A novel method of combining the wavelet and RLS to forecast the Internet traffic is discussed. The focus of this article is how to exploit the correlation structure to make accurate forecast of the Internet traffic, where the property of self-similarity or long-range dependence plays an important role. First, it is shown that through the wavelet transform, the long-range dependence of the temporal network traffic is destructed to short-range dependence among the wavelets. Such short-range dependence can be approximated with a linear correlation structure. Also the approximation coefficients can be fairly well forecast with a linear filter. Then, the method of combining the wavelet and RLS is used to forecast the Internet traffic and is applied to the empirical traffic data from Bellcore. The result shows that our new method achieves extraordinary accuracy.