Techniques for the Analysis of Modern Web Page Traffic using Anonymized TCP/IP Headers Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 22, 2019
  • Sanders, Sean
    • Affiliation: College of Arts and Sciences, Department of Computer Science
  • Analysis of traces of network traffic is a methodology that has been widely adopted for studying the Web for several decades. However, due to recent privacy legislation and increasing adoption of traffic encryption, often only anonymized TCP/IP headers are accessible in traffic traces. For traffic traces to remain useful for analysis, techniques must be developed to glean insight using this limited header information. This dissertation evaluates approaches for classifying individual web page downloads — referred to as web page classification — when only anonymized TCP/IP headers are available. The context in which web page classification is defined and evaluated in this dissertation is different from prior traffic classification methods in three ways. First, the impact of diversity in client platforms (browsers, operating systems, device type, and vantage point) on network traffic is explicitly considered. Second, the challenge of overlapping traffic from multiple web pages is explicitly considered and demultiplexing approaches are evaluated (web page segmentation). And lastly, unlike prior work on traffic classification, four orthogonal labeling schemes are considered (genre-based, device-based, navigation-based, and video streaming-based) — these are of value in several web-related applications, including privacy analysis, user behavior modeling, traffic forecasting, and potentially behavioral ad-targeting. We conduct evaluations using large collections of both synthetically generated data, as well as browsing data from real users. Our analysis shows that the client platform choice has a statistically significant impact on web traffic. It also shows that change point detection methods, a new class of segmentation approach, outperform existing idle time-based methods. Overall, this work establishes that web page classification performance can be improved by: (i) incorporating client platform differences in the feature selection and training methodology, and (ii) utilizing better performing web page segmentation approaches. This research increases the overall awareness on the challenges associated with the analysis of modern web traffic. It shows and advocates for considering real-world factors, such as client platform diversity and overlapping traffic from multiple streams, when developing and evaluating traffic analysis techniques.
Date of publication
Resource type
Rights statement
  • In Copyright
  • Kaur, Jasleen
  • Aikat, Jay
  • Jeffay, Kevin
  • Monrose, Fabian
  • Beyah, Raheem
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill
Graduation year
  • 2017

This work has no parents.