Traffic prediction base on the sampled data

Prediction of network traffic based on the observed data is very important. If we execute that by every packet, it requires large amount of calculation with respect to time, so we need to predict based on the observed data obtained at each sample time. We have developed a new method of filtering and prediction using Kolmogorovpsilas forward equation, for it is very convenient to calculate the conditional distribution explicitly for any time. Although our method focused on Ito diffusion process and some sort of Levy process, the network traffic is known to show self-similarity and long rage dependency, modeled by Fractional Brownian Motion (FBM in the following), whose Hurst parameter H is greater than frac12. In this paper, we adopt our method for FBM with a drift term with a constant coefficient, and Hurst parameter H > frac12.