On the feasibility of profiling internet users based on volume and time of usage

Abstract

In this paper, we address the issue of profiling users over the Internet using meta-data logs derived from network flow data (hence preserving a high degree of privacy). In this broader context, we specifically aim to empirically demonstrate that Internet volume and time of usage of humans do exhibit repeatable behavior over time. In our experimental study, Internet usage statistics of octets and duration (collected via privacy-preserving NetFlow records) of 66 student subjects in a college campus was recorded for a month. Subsequently, using state-of-the-art statistical techniques, we demonstrate how the Internet usage of any particular subject is highly-correlated with usage of the same subject over multiple time scales, while simultaneously being distinct from usage of other subjects. We derive interesting and practically useful trends on the relationship between the degree of distinguishability and the window time chosen to do the profiling. We also present discussions on the practical applications of this new study.