Author

Haitao Du

Abstract

Prevalent computing devices with networking capabilities have become critical network infrastructure for government, industry, academia and every-day life. As their value rises, the motivation driving network attacks on this infrastructure has shifted from the pursuit of notoriety to the pursuit of profit or political gains, leading to network attack on various scales. Facing diverse network attack strategies and overwhelming alters, much work has been devoted to correlate observed malicious events to pre-defined scenarios, attempting to deduce the attack plans based on expert models of how network attacks may transpire.

We started the exploration of characterizing network attacks by investigating how temporal and spatial features of attack sequence can be used to describe different types of attack sources in real data set. Attack sequence models were built from real data set to describe different attack strategies. Based on the probabilistic attack sequence model, attack predictions were made to actively predict next possible actions. Experiments through attack predictions have revealed that sophisticated attackers can employ a number of obfuscation techniques to confuse the alert correlation engine or classifier. Unfortunately, most exiting work treats attack obfuscations by developing ad-hoc fixes to specific obfuscation technique. To this end, we developed an attack modeling framework that enables a systematical analysis of obfuscations.

The proposed framework represents network attack strategies as general finite order Markov models and integrates it with different attack obfuscation models to form probabilistic graphical model models. A set of algorithms is developed to inference the network attack strategies given the models and the observed sequences, which are likely to be obfuscated. The algorithms enable an efficient analysis of the impact of different obfuscation techniques and attack strategies, by determining the expected classification accuracy of the obfuscated sequences. The algorithms are developed by integrating the recursion concept in dynamic programming and the Monte-Carlo method.

The primary contributions of this work include the development of the formal framework and the algorithms to evaluate the impact of attack obfuscations. Several knowledge-driven attack obfuscation models are developed and analyzed to demonstrate the impact of different types of commonly used obfuscation techniques. The framework and algorithms developed in this work can also be applied to other contexts beyond network security. Any behavior sequences that might suffer from noise and require matching to pre-defined models can use this work to recover the most likely original sequence or evaluate quantitatively the expected classification accuracy one can achieve to separate the sequences.

Library of Congress Subject Headings

Computer networks--Security measures; Cyberterrorism--Prevention; Cyberterrorism--Mathematical models

Publication Date

8-2014

Document Type

Dissertation

Student Type

Graduate

Degree Name

Computing and Information Sciences (Ph.D.)

Department, Program, or Center

PhD Program in Computing and Information Sciences

Advisor

Shanchieh Jay Yang

Advisor/Committee Member

Pengcheng Shi

Advisor/Committee Member

Bo Yuan

Campus

RIT – Main Campus

Plan Codes

COMPIS-PHD

Share

COinS