Multi-scale Combination Prediction Model with Least Square Support Vector Machine for Network Traffic

Abstract

A revised multi-scale prediction combination model for network traffic is proposed, where network traffic series are decomposed with stationary wavelet transform, and the different models are built with combinations of wavelet decomposition coefficients. LS-SVM is introduced to predict the coefficients at the expectation point, the prediction value can be obtained by wavelet inversion transform. The simulation experiments with the two traffic traces at different time scale are done with the proposed system, and other predictors. The correlation structure between the prediction point and history data is also explored. The results show that the proposed model improve the computability and achieve a better forecasting accuracy.