Events, Patterns, and Analysis: Forecasting International Conflict in the Twenty-First Century

Project Members



This project is funded in part by an NSF ITR and an equipment grant from CITI for use of the Rice Terascale Cluster.

Our NSF proposal is here .

The Project

Project Context

With the end of the Cold War, the threat of a global-scale nuclear conflagration has receded. Unfortunately, this has not meant that all forms of serious conflict have been eliminated. Focusing only on the United States, we see that since the fall of the Berlin Wall, the United States has engaged in three wars (the Gulf War, Kosovo, and Afghanistan), as well as having its people - and even its territory - subject to attack. Other areas of the world have either experienced even more conflict (the Middle East) or threaten to do so in the immediate future (South Asia, the Korean Peninsula). Needless to say, should any of these conflicts break out and escalate, we might witness (albeit on a small scale) the nuclear conflagration that the US and the Soviet Union managed to avoid.

Project Goals

We believe that the proliferation of news in electronic form as well as a series of advances in information extraction, data mining, statistical machine learning and stochastic modeling have made it possible to predict the outbreak of a serious international conflict by analyzing event data extracted from a multitude of sources over an extended period of time. The goal of our project is to develop techniques to construct extensive event data sets and models necessary to make such predictions. We hope to be able to predict the onset of serious international conflicts four to eight weeks in advance. Specifically, the goals of our research are: Timely warning of the outbreak of serious conflict can be a key element in conflict resolution. Early warning can provide the time for state and non-state actors to intervene and prevent the outbreak. Thus, we feel our work can be of potential value to the conflict resolution process, even though the focus of our research is predicting the outbreak and evolution of conflict.

Preliminary results

Gathering Relevant News Stories
Generating Events Data from News Stories
Analyzing Events Data
At this stage, we have derived results using existing events data collections (since the process of gathering new events is in process, as indicated above). We have used a wavelet-based protocol to demonstrate that serious international conflicts reveal themselves as singularities in the events data, and that such singularities are often predictable 6-8 weeks in advance of their occurrence. We have validated the method in the context of events data extracted from 20 years (1979-1999) of events data gathered from the Reuters articles on the Middle East, as well as an existing collection of events pertaining to the Cold War We have published these results in the paper referenced below. For a taste of our analysis, see this graph for a wavelet analysis of event data from the Middle East for the period 1979-2001. The modulus maxima lines correspond to three major wars in the region during this period If conflicts between Iran and Iraq are removed from the event data, we can see the skirmishes between other states in the region during this period.


Events, Patterns, and Analysis: Forecasting International Conflict in the Twenty-First Century", R. Stoll and D. Subramanian, in Robert Trappl (ed.) Computer-Aided Methods for International Conflict Resolution and Prevention Springer Verlag, (in press) (presentation version available from here )

Core Scientific Question

How well can an objective, data-driven approach to modeling the genesis and evolution of conflict in various regions of the world work?

Our ultimate goal

To change the science of conflict prediction in a fundamental way by exploiting the electronic proliferation of information on political events, and by harnessing available computing power to comprehensively and objectively analyze international conflicts.