Forecasting Network Traffic at Large Time Scale by Using Dual-related Method

The accuracy of network traffic prediction has received significant interest in various domains, such as capacity planning, anomaly detection, admission control, and traffic engineering. For large-time scale traffic variation, it shows both a daily pattern and an hour pattern, which means the model based on single trend has not met the needs of prediction. Therefore, by dealing with the internal relationship of large-time scale network traffic, this paper combines the regular trend and the smooth or seasonal trend of hours and days, then fit the dual-related model to predict large-time scale traffic. The result indicates that the proposed model effectively identified the correlations of data between days and hours, and is successful in forecasting approaches.