Improving Self Organizing Map Performance for Network Intrusion Detection

The continuous evolution of the types of attacks against computer networks suggests a paradigmatic shift from misuse based intrusion detection system to anomaly based systems. Unsupervised learning algorithms are natural candidates for this task, but while they have been successfully applied in host-based intrusion detection, network-based applications are more difficult, for a variety of reasons, including performance. We propose an architecture which implements a network-based, anomaly based intrusion detection system, which uses unsupervised learning algorithms. In this paper we describe the improvements and modifications needed in order to increase the throughput of a Self Organizing Map algorithm and make it able to handle high dimensional input data at a rate suitable for Intrusion Detection purposes at