Machine Learning and Civil War: Investigating Tree-Based Models for Predicting Intrastate Violence
Embargo until
Date
2017-12
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This study’s aim is to improve the forecasting of civil war and examine the practical utility of using machine learning techniques in this effort. Specifically, this study investigates a variety of sampling methods used to construct useful models from imbalanced data, the algorithm used to construct these models, and which of the models built by previous scholars is the most useful for prediction when different sampling procedures algorithms are applied. This study finds that up-sampling and SMOTE sampling generally improve model performance, that tree-based ensemble methods generally perform significantly better than logistic regression and that of these ensemble methods Extreme Gradient Boosting generally performs the best, and that the previous model constructed by Collier & Hoeffler performs extremely well, especially when combined with sampling procedures and tree-based ensemble methods.
Description
Keywords
civil war, machine learning, decision trees, intrastate violence