Violence and Cell Phone Communication: Behavior and Prediction in Cote D’Ivoire

Abstract

We investigate the relationship between low-level incidents of political violence and communication patterns seven months after the 2012 Ivorian Civil War using network traffic from all of Orange Telecom’s Cote d’Ivoire cell towers and 500,000 randomly sampled cell phone subscribers. We first show that in the days preceding small violent incidents, mobile phone call volumes increase by 10% and the number of cell phones that are active increases by 6%, while the length of average calls decreases by 4%. These unique communication patterns attenuate as the distance from violence increase and strengthen when incidents with no fatalities are excluded. We then use machine learning techniques to explore whether these changes can predict the day and location of violent events. The addition of cell phone data to base models appear to improve our ability to predict violent events at very fine spatial and temporal resolution.

Keywords: prediction, information aggregation, communication, violence, machine learning, conflict prevention, big data, information and communications technology

JEL Classification: C53, D74 D83, D84, N37, C55, Q34