Date of Award

3-22-2012

Document Type

Thesis

Degree Name

Master of Science

Department

Department of Electrical and Computer Engineering

First Advisor

Gary B. Lamont, PhD.

Abstract

Detecting attacks targeted against military and commercial computer networks is a crucial element in the domain of cyberwarfare. The traditional method of signature-based intrusion detection is a primary mechanism to alert administrators to malicious activity. However, signature-based methods are not capable of detecting new or novel attacks. This research continues the development of a novel simulated, multiagent, flow-based intrusion detection system called MFIRE. Agents in the network are trained to recognize common attacks, and they share data with other agents to improve the overall effectiveness of the system. A Support Vector Machine (SVM) is the primary classifier with which agents determine an attack is occurring. Agents are prompted to move to different locations within the network to find better vantage points, and two methods for achieving this are developed. One uses a centralized reputation-based model, and the other uses a decentralized model optimized with stochastic search. The latter is tested for basic functionality. The reputation model is extensively tested in two configurations and results show that it is significantly superior to a system with non-moving agents. The resulting system, MFIRE-2, demonstrates exciting new network defense capabilities, and should be considered for implementation in future cyberwarfare applications.

AFIT Designator

AFIT-GCO-ENG-12-12

DTIC Accession Number

ADA557998

Comments

Detecting attacks targeted against military and commercial computer networks is a crucial element in the domain of cyberwarfare. The traditional method of signature-based intrusion detection is a primary mechanism to alert administrators to malicious activity. However, signature-based methods are not capable of detecting new or novel attacks. This research continues the development of a novel simulated, multiagent, flow-based intrusion detection system called MFIRE. Agents in the network are trained to recognize common attacks, and they share data with other agents to improve the overall effectiveness of the system. A Support Vector Machine (SVM) is the primary classifier with which agents determine an attack is occurring. Agents are prompted to move to different locations within the network to find better vantage points, and two methods for achieving this are developed. One uses a centralized reputation-based model, and the other uses a decentralized model optimized with stochastic search. The latter is tested for basic functionality. The reputation model is extensively tested in two configurations and results show that it is significantly superior to a system with non-moving agents. The resulting system, MFIRE-2, demonstrates exciting new network defense capabilities, and should be considered for implementation in future cyberwarfare applications.

Share

COinS