Beta-hebbian learning for visualizing intrusions in flows

Abstract

The present research work focuses on Intrusion Detection (ID), identifying “anomalous” patterns that may be related to an attack to a system or a network. In order to detect such anomalies, this present paper proposes the visualization of network flows for ID by applying a novel neural method called Beta Hebbian Learning (BHL). Four real-life traffic segments from the University of Twente datasets have been analysed by means of the BHL. Such datasets were gathered from a honeypot directly connected to the Internet so it is guaranteed that it contains real-attack data. Results obtained by BHL provide clear evidence of the ID System clearly separating the different types of attacks present in each dataset and outperforming other well-known projection algorithms.