The New Jersey Institute of Technology's
Electronic Theses & Dissertations Project

Title:	Statistical anomaly denial of service and reconnaissance intrusion detection
Author:	Zhang, Zheng
View Online:	njit-etd2004-106 (xvi, 118 pages ~ 5.4 MB pdf)
Department:	Department of Electrical and Computer Engineering
Degree:	Doctor of Philosophy
Program:	Electrical Engineering
Document Type:	Dissertation
Advisory Committee:	Manikopoulos, Constantine N. (Committee chair) Dhawan, Atam P. (Committee member) Ziavras, Sotirios (Committee member) Rojas-Cessa, Roberto (Committee member) Ucles, Jose Luis (Committee member)
Date:	2004-05
Keywords:	Intrusion detection Neural networks Reconnaisance attacks PDF statistics Denial-of-service attacks HIDE
Availability:	Unrestricted
Abstract:	This dissertation presents the architecture, methods and results of the Hierarchical Intrusion Detection Engine (HIDE) and the Reconnaissance Intrusion Detection System (RIDS); the former is denial-of-service (DoS) attack detector while the latter is a scan and probe (P&S) reconnaissance detector; both are statistical anomaly systems. The HIDE is a packet-oriented, observation-window using, hierarchical, multi-tier, anomaly based network intrusion detection system, which monitors several network traffic parameters simultaneously, constructs a 64-bin probability density function (PDF) for each, statistically compares it to a reference PDF of normal behavior using a similarity metric, then combines the results into an anomaly status vector that is classified by a neural network classifier. Three different data sets have been utilized to test the performance of HIDE; they are OPNET simulation data, DARPA'98 intrusion detection evaluation data and the CONEX TESTBED attack data. The results showed that HIDE can reliably detect DoS attacks with high accuracy and very low false alarm rates on all data sets. In particular, the investigation using the DARPA'98 data set yielded an overall total misclassification rate of 0.13%, false negative rate of 1.42%, and false positive rate of 0.090%; the latter implies a rate of only about 2.6 false alarms per day. The RIDS is a session oriented, statistical tool, that relies on training to model the parameters of its algorithms, capable of detecting even distributed stealthy reconnaissance attacks. It consists of two main functional modules or stages: the Reconnaissance Activity Profiler (RAP) and the Reconnaissance Alert Correlater (RAC). The RAP is a session-oriented module capable of detecting stealthy scanning and probing attacks, while the RAG is an alert-correlation module that fuses the RAP alerts into attack scenarios and discovers the distributed stealthy attack scenarios. RIDS has been evaluated against two data sets: (a) the DARPA'98 data, and (b) 3 weeks of experimental data generated using the CONEX TESTBED network. The RIDS has demonstrably achieved remarkable success; the false positive, false negative and misclassification rates found are low, less than 0.1%, for most reconnaissance attacks; they rise to about 6% for distributed highly stealthy attacks; the latter is a most challenging type of attack, which has been difficult to detect effectively until now.