Machine Learning and Security Studies
I present three papers that demonstrate how the cultural and technological advances in the machine learning literature can and should impact quantitative security studies. Although the math behind many machine learning techniques is not dissimilar to many statistical techniques, machine learning is heavily focused on predictive capacity. I argue that quantitative security studies should take a similar approach and take advantage of the advances made in the machine learning literature. In the first paper I analyze the practical utility of economic shocks as a tool for forecasting violence in four African countries. Several similar studies have published statistically significant findings that carry implicit or explicit recommendations to policy makers. However, I find that despite the technical sophistication of the original papers, the mechanism fails to add value to otherwise similar forecasting models, casting serious doubt on the actual utility of similar models. In the second paper I present PEMA: a flexible and powerful new system for creating event data using machine learning. Stagnation in the data available for quantitative analysis frustrates progress and can lead to erroneous findings and PEMA can play an import role in addressing that problem. In the final paper I argue that digital insecurities reduce conflict and encourage cooperation by decreasing the amount of private information between states. Testing my argument requires dealing with complex systems and data for which common statistical tests would be inadequate. Again, I turn to machine learning to show that digital information can play a strong role in encouraging interstate cooperation.
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 United States License.
Rights for Collection: Duke Dissertations