NIDS: False Positive Reduction Through Anomaly Detection presented at BlackhatUSA 2006

by Emmanuele Zambon, Damiano Bolzoni,

Summary : The Achilles' heel of network IDSs lies in the large number of false positives (i.e., false attacks) that occur: practitioners as well as researchers observe that it is common for a NIDS to raise thousands of mostly false alerts per day. False positives are a universal problem as they affect both signature-based and anomaly-based IDSs. Finally, attackers can overload IT personnel by forging ad-hoc packets to produce false alerts, thereby lowering the defences of the IT infrastructure.
Our thesis is that one of the main reasons why NIDSs show a high false positive rate is that they do not correlate input with output traffic: by observing the output determined by the alert-raising input traffic, one is capable of reducing the number of false positives in an effective manner. To demonstrate this, we have developed APHRODITE (Architecture for false Positives Reduction): an innovative architecture for reducing the false positive rate of any NIDS (be it signature-based or anomaly-based). APHRODITE consists of an Output Anomaly Detector (OAD) and a correlation engine; in addition, APHRODITE assumes the presence of a NIDS on the input of the system. For the OAD we developed POSEIDON (Payl Over Som for Intrusion DetectiON): a two-tier network intrusion detection architecture.
Benchmarks performed on POSEIDON and APHRODITE with DARPA 1999 dataset and with traffic dumped from a real-world public network show the effectiveness of the two systems. APHRODITE is able to reduce the rate of false alarms from 50% to 100% (improving accuracy) without reducing the NIDS ability to detect attacks (completeness).