Combined Coiflet wavelet and recursive least squares predictor for MPEG traffic prediction

It has been shown both theoretically and empirically that network traffic has a self similar nature. The performance evaluation parameters and QoS metrics depend on this self similar nature of network traffic. So the network traffic predictor can be vital in applications such as bandwidth allocation and network capacity planning. A new wavelet based predictor in combination with adopted recursive least squares (RLS) is presented in this paper. We show that using 2 different wavelets yields different mean absolute errors in prediction of the self similar MPEG traffic. Previous work uses Daubechies 40 wavelet in combination with RES. Our experimental results show that using the Coiflet 5 in comparison to Daubechies 40 used by X. Wang has a mean absolute error that is 5.68 orders of magnitude less than Daubechies 40.