Suppression of Boundary Effect and Introduction of Scale Correlation for Wavelet based Traffic Prediction

Abstract

In this paper, we propose a Wavelet-based prediction method of the Internet traffic volume. By introducing maximal overlap formulation of the Haar wavelet, the proposed method is free from so-called boundary condition, which arises from processing delay of analysis filters of the wavelet transform and refrains from full utilization of information on recent input signals. The proposed method is based on a vector autoregressive model so as to introduce inter-scale correlation of Wavelet coefficient series at different scales.