A hybrid method to intrusion detection systems using HMM

Abstract

IDS use different sources of observation data and a variety of techniques to differentiate between benign and malicious behaviors. In the current work, Hidden Markov Models (HMM) are used in a manner analogous to their use in text categorization. The proposed approach performs host-based intrusion detection by using HMM along with STIDE methodology (enumeration of subsequences) in a hybrid fashion. The proposed method differs from STIDE in that only one profile is created for the normal behavior of all applications using short sequences of system calls issued by the normal runs of the programs. Subsequent to this, HMM with simple states along with STIDE is used to categorize an unknown program's sequence of system calls to be either normal or an intrusion. The results on 1998 DARPA data show that the hybrid method results in low false positive rate with high detection rate. © Springer-Verlag Berlin Heidelberg 2005.