Prediction for Long Range Bursty Traffic

Fractional AutoRegressive Integrated Moving Averag (FARIMA) processes with alpha-stable innovation can capture non-Gaussian, namely heavy tailness that is the key factor of the long range burstiness of self-similar traffic. A FARIMA process can be regarded as ARMA process driven by fractional differencing (FD) process. ARMA processes with infinite variance can be simulated with recurrent neural network (RNN) instead of conventional Least Squares methods. We adopt three intelligent methods to train the weights of RNN in order to minimize the dispersion. The final predicted values are combined previous three predicted values. Prediction experimental results for the actual traffic trace show that the three FARIMA predictors are efficient, the compound predictors are more accurate.