The Geography of Anonymous Communications: Predicting Escalation of Anonymity Networks During Events of Civil Unrest

Abstract

Civil unrest can trigger uptake of anonymous communication to protect user identity and location or to circumvent censorship. Anonymity networks such as the Tor Network can support planning, orchestrating, or responding to protest events. This research aimed to understand this relationship between protest events and Tor usage. A methodology was developed to discover the best supervised learning method for predicting Tor usage in response to protest events. Twelve classification algorithms were evaluated using data representing the different conflict event types, number of events per day, source and target actor categories, fatalities, and Tor usage.  Experiments were conducted using over five years of conflict event data from nine countries selected from the Africa-based Armed Conflict Location and Event Dataset (ACLED). This research produced unique quantitative results predicting Tor escalation during conflict events with an F1 Score of over 86%. Results are significant given the multitude of use cases for Tor, with the strongest signal occurring in authoritarian regimes.