Nonlinear-periodical network traffic behavioral forecast based on seasonal neural network model

How to predict Internet behavior is a standing challenge to the study of network behavior. Traditionally, the ARIMA model, always used for network traffic prediction, has difficulties in deciding its parameter values and, therefore, finds it hard to deal with the condition of a nonlinear time series. The paper presents a seasonal neural network prediction model for monitoring network traffic based on the neural network model of a time series and the periodic trend of traffic behavior. In addition, a series of data processes are taken to improve the prediction accuracy. The result of the application of the model to CERNET traffic prediction shows the model´s reasonableness, and that it is more accurate than the ARIMA model and the neural network common time series.