Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Author ORCID Identifier


Open Access Dissertation

Document Type


Degree Name

Doctor of Philosophy (PhD)

Degree Program

Computer Science

Year Degree Awarded


Month Degree Awarded


First Advisor

Amir Houmansadr

Second Advisor

Donald F. Towsley

Third Advisor

Phillipa Gill

Fourth Advisor

Dennis L. Goeckel


Traffic analysis is the practice of using inherent characteristics of a network flow such as timings, sizes, and orderings of the packets to derive sensitive information about it. Traffic analysis techniques are used because of the extensive adoption of encryption and content-obfuscation mechanisms, making it impossible to infer any information about the flows by analyzing their content. In this thesis, we use traffic analysis to infer sensitive information for different objectives and different applications. Specifically, we investigate various applications: p2p cryptocurrencies, flow correlation, and messaging applications. Our goal is to tailor specific traffic analysis algorithms that best capture network traffic’s intrinsic characteristics in those applications for each of these applications. Also, the objective of traffic analysis is different for each of these applications. Specifically, in Bitcoin, our goal is to evaluate Bitcoin traffic’s resilience to blocking by powerful entities such as governments and ISPs. Bitcoin and similar cryptocurrencies play an important role in electronic commerce and other trust-based distributed systems because of their significant advantage over traditional currencies, including open access to global e-commerce. Therefore, it is essential to the consumers and the industry to have reliable access to their Bitcoin assets. We also examine stepping stone attacks for flow correlation. A stepping stone is a host that an attacker uses to relay her traffic to hide her identity. We introduce two fingerprinting systems, TagIt and FINN. TagIt embeds a secret fingerprint into the flows by moving the packets to specific time intervals. However, FINN utilizes DNNs to embed the fingerprint by changing the inter-packet delays (IPDs) in the flow. In messaging applications, we analyze the WhatsApp messaging service to determine if traffic leaks any sensitive information such as members’ identity in a particular conversation to the adversaries who watch their encrypted traffic. These messaging applications’ privacy is essential because these services provide an environment to dis- cuss politically sensitive subjects, making them a target to government surveillance and censorship in totalitarian countries. We take two technical approaches to design our traffic analysis techniques. The increasing use of DNN-based classifiers inspires our first direction: we train DNN classifiers to perform some specific traffic analysis task. Our second approach is to inspect and model the shape of traffic in the target application and design a statistical classifier for the expected shape of traffic. DNN- based methods are useful when the network is complex, and the traffic’s underlying noise is not linear. Also, these models do not need a meticulous analysis to extract the features. However, deep learning techniques need a vast amount of training data to work well. Therefore, they are not beneficial when there is insufficient data avail- able to train a generalized model. On the other hand, statistical methods have the advantage that they do not have training overhead.