Predicting Violence in Sudan for Humanitarian Relief

This paper from the Systems and Information Engineering Design Symposium (SIEDS) 2014 conference builds a series of models using GDELT to predict future violent unrest in Sudan with 80% accuracy.

As part of their study, the team manually reviewed 50 randomly selected Sudan and South Sudan events from GDELT and compared the details in the event record against the event description in the source article, finding very close alignment:

  • Actor1: 79% correct
  • Actor2: 78.5% correct
  • Geography: 75.4%
  • Event Code: 81.2%

Given that both Sudan and South Sudan have few international correspondents on the ground, news coverage often comes from a wide-ranging array of international, national, regional, and local sources, so the high level of accuracy that GDELT demonstrated in correctly codifying the contents of those news reports is particularly encouraging.

Read the Full Conference paper.