anhqle / conflict_prediction

Predicting conflict using machine learning

Conflict Prediction with Machine Learning

Through Ward Lab at Duke University, this project is part of the Integrated Conflict Early Warning System (ICEWS), a Defense Advanced Research Projects Agency's (DARPA) program. The goal of ICEWS was to develop and integrate multiple data feeds and analytic models to generate reliable forecasts of violence and upheaval around the world.

Tasks performed:

• Predicted world-wide conflicts using country-month data (20,000 rows × 500 variables), achieving over 95% on precision and recall, outperforming all current models
• Used parallel R to rapidly fit Bayesian variable selection and boosted decision tree, leading to a 4 times speed-up

To do:

• Adapt the model for insurgency to predict mass protest