Predicting the Spread of Terrorist Organizations Using Graphs
Abstract
The U.S. Defense and Intelligence communities expend vast amounts of resources tracking and trying to predict the geographic spread of terrorist groups such as the Islamic State of Iraq and Syria (ISIS). Current approaches to this problem use a variety of social, demographic, and geographic data to make predictions about the spread of a terrorist organization. We demonstrate a novel approach that converts the geographic area of interest, Iraq and Syria, into a graph with the populated places as nodes and the road network as the edges of the graph. We then use this graph to compute graph-based statistics such as measures of centrality and first-order neighbor statistics on the nodes in the graph. By adding the graph-based features, we combine social, demographic, and geographic data with data that quantifies the relationships between the populated places in Iraq and Syria. This ultimately improves predictive performance for predicting future territorial gains and losses by ISIS. Furthermore, our models demonstrate that the graph-based features are the most influential variables in predicting whether or not a node will be in or out of ISIS territory.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2018
- Accession Number
- AD1060093
Entities
People
- Anthony B. Vanderzee
Organizations
- Naval Postgraduate School