Measuring Conflict Between Countries from a Data-Driven Perspective

Abstract

Research on international conflict has previously focused on explaining events in terms of their stages such as the onset or escalation of major crises. Recently, however, global conflict modeling has received growing attention. Predictions of violent events, in particular, have been increasingly accurate using various methods ranging from expert knowledge to quantitative methods and conflict forecasting systems. The DARPA Integrated Crisis Early Warning System (ICEWS) is an effort to apply computational social science models to dynamic data sources to produce a coherent event dataset that enables actionable forecasts of foreign country stability. While international crisis forecasting is becoming increasingly prevalent using the ICEWS event data, little research exists on the development of feature optimization techniques to train machine learning-enabled conflict forecasting systems. In this project, we propose a methodology to predict future conflict behavior through feature optimization with the ICEWS event data. We performed various time-series analyses on the ICEWS event data in order to develop features from its most meaningful crisis indicators. Through an investigation of Russian conflict behavior patterns with former Soviet Union countries, we used machine learning to produce highly accurate forecasts for Russian crises one month in advance. For future work, we hope to create a more generalized early crisis detection system using data from every country in the ICEWS event dataset.