Combinatorial Methodology Study for Network Traffic Prediction

Network traffic is a typical time-series data with strong lag and aftereffect, for the existence of local optimum, time-consuming and other defects in the method for the currently determining number of lags, this paper presents a combination of network traffic prediction method (GS-GA-LSSVM). At first, using geo-statistics (GS) to quickly determine the optimal lag order of network traffic, then reconstructing network traffic with the lag order, and finally using genetic algorithm (GA) to optimize least squares supporting vector machine (LSSVM) to model predictions for network traffic. Simulation results show that, GS-GA-LSSVM on network traffic prediction accuracy is better than any of the participating models, and which can better reflect the complex dynamics of network traffic disciplines.