Anomaly Detection in Network Traffic Using Selected Methods of Time Series Analysis

Abstract

In this paper a few methods for anomaly detection in computer networks with the use of time series methods are proposed. The special interest was put on Brown's exponential smoothing, seasonal decomposition, naive forecasting and Exponential Moving Average method. The validation of the anomaly detection methods has been performed using experimental data sets and statistical analysis which has shown that proposed methods can efficiently detect unusual situations in network traffic. This means that time series methods can be successfully used to model and predict a traffic in computer networks as well as to detect some unusual or unrequired events in network traffic.