Network traffic prediction based on multi-scale wavelet transform and mixed time series model

Focusing on the character of long time correlation of network traffic data, a hybrid model based on multi-scale wavelet transform, ARMA model and ARFIMA model is proposed. The original data are transferred into four layers data by Mallat algorithm, and ARMA models apply in approximate layers data to predict the future trend, and ARFIMA model apply in detail layers data to predict the future volatility, then we reconstruct them into predicted the network data. The simulation experiment on the hybrid model is conduced by using the data collected from the university network system. The experiment result shows that the hybrid ARMA model and ARFIMA model has higher accuracy on predication the network traffic and is practical on network management and optimization.