Non-Stationary Markov Models and Anomaly Propagation Analysis in IDS

We propose an anomaly based IDS that results in a decreased rate of false positives. It employs the new means of host-based detection in the system call domain with correlating anomalies reported by different hosts to the IDS server. A novel anomaly detection mechanism operating at the host level treats an application or service as a non-stationary stochastic process and models it as a non- stationary Markov chain that significantly improves model accuracy. A server-based procedure for the detection of anomaly propagation is employed. While false alarms do not propagate within the network, detected anomaly propagation with a high degree of certainty can be attributed to a computer worm; otherwise the alarms are to be treated as false positives.