Statistical methods used for intrusion detection

Abstract

Computer networks are being attacked everyday. Intrusion detection systems are used to detect and reduce effects of these attacks. Signature based intrusion detection systems can only identify known attacks and are ineffective against novel and unknown attacks. Intrusion detection using anomaly detection aims to detect unknown attacks and there exist algorithms developed for this goal. In this study, performance of five anomaly detection algorithms and a signature based intrusion detection system is demonstrated on synthetic and real data sets. A portion of attacks are detected using Snort and SPADE algorithms. PHAD and other algorithms could not detect considerable portion of the attacks in tests due to lack of sufficiently long enough training data.