Internet Traffic Forecasting using Neural Networks

Abstract

The forecast of Internet traffic is an important issue that has received few attention from the computer networks field. By improving this task, efficient traffic engineering and anomaly detection tools can be created, resulting in economic gains from better resource management. This paper presents a Neural Network Ensemble (NNE) for the prediction of TCP/IP traffic using a Time Series Forecasting (TSF) point of view. Several experiments were devised by considering real-world data from two large Internet Service Providers. In addition, different time scales (e.g. every five minutes and hourly) and forecasting horizons were analyzed. Overall, the NNE approach is competitive when compared with other TSF methods (e.g. Holt-Winters and ARIMA).Paulo Cortez is with the

A prediction criterion for selecting popular destinations

“…More complex predictors could also be employed in our study, such as Auto-Regressive Integrated Moving-Average (ARIMA) (that combines linearly past traffic volumes and/or errors) [16] and Neuronal Networks (NN) (here the basic idea is to train a NN with past traffic volumes to predict future values) [17]. We only employed LV, MA and LpEMA in the belief that there is no advantage in using complex predictors given the fact that the performance achieved is almost the same as with the simpler predictors [18], [19].…”

Improving the Traffic Prediction Capability of Neural Networks Using Sliding Window and Multi-task Learning Mechanisms

“…In order to solve this problem, we used several instances of each specific model, sharing the same scenario but having different internal initial conditions. This solution is called Neural Network Ensemble -Injecting Randomness [9]. The final prediction is given by the average of the best (where the mean error of the last C iterations is less than a threshold CT ) individual predictions [33].…”

“…Using a set of these models, each NN model dedicated to a specific network link, the network manager can have an important tool to anticipate network upgrade decisions or changes on the network functional operation, like for example routing decisions. We know that traffic prediction can be used on MPLS (Multiprotocol Label Switching) based networks to optimize the set of virtual circuits that needs to be established, while for non-MPLS networks the weights associated to the routing protocols can be defined in a more efficient way using the knowledge of the future traffic demand on each network link [9]. Besides, network traffic forecasting can help on the detection of some security anomalies [10]; typical security attacks, such as Denial-of-Service (DoS), can be automatically detected by performing an on-line analysis/comparison between the expected amount or type of traffic and the current measured data in the entire network or in some specific link.…”

Topology Aware Internet Traffic Forecasting Using Neural Networks

“…Hence, it is important to develop techniques to better understand and forecast the behavior of these networks. In effect, TCP/IP traffic prediction is gaining more attention from the computer networks community [18,12,1,2]. By improving this task, network providers can optimize resources, allowing a better quality of service.…”