MSc thesis defense School of Computer Science-Þorgeir Auðunn Karlsson

Epidemiological surveillance through cellphone metadata

  • 28.5.2018, 14:30 - 15:30

Title: Epidemiological surveillance through cellphone metadata
Author: Þorgeir Auðunn Karlsson
Date and Time: May 28, 2018 at 14:30
Location: M104

Supervisor: Dr. Ýmir Vigfússon Reykjavík University 

Abstract

Epidemics cause significant tolls financially and on the population. A pandemic can originate anywhere on the planet, and global epidemiological surveillance relies crucially on regional infrastructure whose capabilities and qualities differ across countries. Effective outbreak containment and eradication requires quick detection, which increases the need for a surveillance system that uses more common infrastructure. Here, we look to billing records of cell phones as a near-ubiquitous source of passively-collected metadata of real time behavioral information and ask if a population’s interaction with the cell network can be used to quantify the number of symptomatic individuals in a community. An Icelandic mobile network operator provided call-detail records during the 2009 H1N1 outbreak in Iceland. Part I of this thesis determines if the passively collected cell phone metadata can quantifiably capture the behavioral change associated with the symptoms of influenza. Results show that diagnosed individuals move less, talk longer, and initiate fewer phone calls in the few days around their diagnosis. Part II of this thesis leverages the analysis work done in part I and poses the problem of classifying symptomatic individuals in a population with machine learning. We compare four different machine learning models for detection and explore various input representations. Results indicate that there is a sufficient signal in the data for the model to estimate an epidemics curve on an aggregate level.