|
Abstract:
|
Intrusion detection has become a basic infrastructure to guarantee the security of most internetworking applications. With more internetworking applications in the Internet nowadays, enormous volume of audit trails are produced for the analysis within intrusion detection. For this reason, it is critical to reduce the detection computation of intrusion detection to
meet the realtime detection requirement. In this paper, using a formal intrusion detection framework, we propose a new concept of variable-length signature, along with feature selection, to compress the behavior models of our intrusion detection system, USAID[7], that achieves promising detection performance. Intuitively, compact behavior models will make the detection process computationally much cheaper. Our experimental results show that the proposed technique will degrade the detection rate of unknown intrusions, and fortunately, that it achieves a high detection rate for known intrusions with a significantly reduced false alarm rate. As a result, compared to USAID, the size of the behavior model is decreased by 99.52%, and the detection computation is cut down by 81.15% at least. |
