Title: The Global Conflict Risk Index: Artificial intelligence for conflict prevention
Publisher: Publications Office of the European Union
Publication Year: 2019
JRC N°: JRC118746
ISBN: 978-92-76-14306-2 (online)
ISSN: 1831-9424 (online)
Other Identifiers: EUR 30028 EN
OP KJ-NA-30028-EN-N (online)
URI: https://publications.jrc.ec.europa.eu/repository/handle/JRC118746
DOI: 10.2760/004232
Type: EUR - Scientific and Technical Research Reports
Abstract: The Global Conflict Risk Index (GCRI), which was designed by the European Commission’s Joint Research Centre (JRC), is the quantitative starting point of the EU’s conflict Early Warning System. Taking into consideration the needs of policy-makers to prioritize actions towards conflict prevention, the GCRI expresses the statistical risk of violent conflict in a given country in the upcoming one to four years. It is based on open source data and grounded in the assumption that the occurrence of conflict is linked to structural conditions, which are used to compute the probability and intensity of conflicts. While the initial GCRI model was estimated by means of linear and logistic regression models, this report presents a new GCRI model based on the Artificial Intelligence (AI) random forest (RF) approach. The models’ hyperparameters are optimized using a ten-fold cross validation. Overall, it is demonstrated that the random forest GCRI models are internally stable, not overfitting, and have a good predictive power. The precision and accuracy metrics are above 98%, both for the national power and subnational power conflict models. The AI GCRI, as a supplementary modelling method for the European conflict prevention policy agenda, is scientifically robust as a baseline quantitative evaluation of armed conflict risk additional to the linear and logistic regression GCRI.
JRC Directorate:Space, Security and Migration

Files in This Item:
File Description SizeFormat 
ai_gcri_technical_report.pdf1.64 MBAdobe PDFView/Open

Items in repository are protected by copyright, with all rights reserved, unless otherwise indicated.