Students Use AWS to Predict Ebola Outbreaks

This past summer, the United States Geospatial Intelligence Foundation hosted the first Geospatial Intelligence (GEOINT) Hackathon. The goal of the hackathon was to bring together and introduce both non-GEOINT and GEOINT-savvy coders and data scientists to interesting problems requiring inventive coding solutions. In addition to enabling participation from the non-GEOINT coding world, the end result was a working code base that performs a specifically requested set of functions or provides answers as outputs.

Students competed along with professional developers and data scientists. The competitors were challenged to predict where Ebola outbreaks might occur and determine why certain areas of West Africa were not affected. The goal was to develop a solution that could be modified to a new set of conditions and be used by other teams.

Out of 30 participants, a student team named “Team Intern” took first place earning a $15,000 award and free admission to the GEOINT conference that occurred in Washington D.C. They developed a predictive analysis model that revealed a likely pathway for Ebola outbreaks. By using an open-source Python library, the team modeled the spread of disease as it’s carried by contagious people through a network of nodes and edges using network theory. Simply put, Team Intern’s library aimed to capture where sick people travel and why.

Problem

To begin, the team looked at what had been done, the data available, and what they could do to fix the problem within 46 hours.

With the spread of disease on the rise, the population of West Africa in danger, and limited hours, the team had to take in data regarding the fatality rate, immunity rate, average travel distance, transmission rate, as well as geo-referenced statistics to determine the virus movement. Then they developed a model that predicts where Ebola will spread and how many people it will affect based on how contagious people travel.

Since disease control measures at water ports and airports are required to prevent the spread of the disease, the only form of travel was by road. However, more data was created by the options available for each contagious person, such as whether they leave or stay home and where they travel— East, West, North, or South. And once they left home, it was assumed they would be more likely to go to highly populated areas near hospitals in cities, thereby infecting more people.

All of this data was used to create a network theory outlining who was susceptible, who was infected, who was recovered, and where they traveled.

Solution

Team Intern turned to AWS to create a model that employed multiple data sources to predict outbreaks and epidemics. The connections between the susceptible and the infected could chart the spread of the disease at each time stamp and how quickly it would spread based on where the contagious people traveled.

The probability density map mapping nodes and edges was able to predict the spread of disease and model the outbreak based on the algorithm created with the data processed.

Hear directly from Team Intern about the problem and their approach to solving it in this on-demand webinar. Watch the webinar here.

Team “Intern”- R. Blair Mason (U.S. Naval Academy ’16), Briana Neuberger (Rochester Institute of Technology ’16), Dan Simon Rochester Institute of Technology ’16 and Paul Warren (Stanford ’17).