ARIMA model for network traffic prediction and anomaly detection

Abstract

This paper presents the use of a basic ARIMA model for network traffic prediction and anomaly detection. Accurate network traffic modeling and prediction are important for network provisioning and problem diagnosis, but network traffic is highly dynamic. To achieve better modeling and prediction it is needed to isolate anomalies from normal traffic variation. Thus, we decompose traffic signals into two parts normal variations, that follow certain law and are predictable and, anomalies that consist of sudden changes and are not predictable. ARIMA analysis and modeling for network traffic prediction is able to detect and identify volume anomaly or outliers.