Monitoring patterns of violence with the ICRC

SDSC Team:
Share this post


The International Committee of the Red Cross (ICRC) is an independent and neutral organization operating worldwide to help people affected by conflicts and armed violence and to protect victims of war. Knowing whether an humanitarian action is effective requires feedback loops. The ICRC currently relies on observations, interviews and status reports compiled by field analysts to monitor its work. This has several shortcomings, like (i) subjectiveness, (ii) unstructured reports in the form of a text (making it difficult to extract information) and (iii) lack of  human resources to monitor episodes of armed violence across entire countries.


To address these issues, we propose incorporating large amounts of data, quantitative measures of conflict intensity and automated, machine-based event analysis. We process many textual reports on conflicts to uncover patterns of violence in the affected areas. To achieve this, we designed an open-source machine learning approach that can transform unstructured textual events into specific forms of violence, essentially identifying "who-did-what-to-whom". This approach has proven to be both flexible and low resource, as we do not rely on large, labeled datasets.


The results of the model’s classification are matched against the ICRC’s data on its protection work that aims to influence armed forces and groups to fight in accordance with international humanitarian law (IHL). This has allowed the ICRC to monitor the impact of its actions and facilitate more informed, data-driven decision-making in planning future actions.

kidnapping events in Mali


The project was funded by the ETH4DHumanitarian Action Challenge and grew out of a collaboration with Daniel Gatica-Perez (EPFL and IDIAP) and Niklas Stoehr (ETH Zurich). We would like to thank the team at the ICRC, Fiona Terry, David Wanstall, Fabien Dany, Chiara Debenedetti and Aminata Gueye for feedback and discussions that motivated this project.


More case studies


Repositioning drugs for a rare disease

Harnessing the power of deep learning to repurpose drugs for the treatment of the Sanfilippo syndrome.
Private sector

Job title standardization through entity alignment of knowledge graphs

Unifying heterogeneous knowledge graphs by applying Natural Language Models and Graph Theory in order to standardize job titles.
Public Sector

Modelling the end-user Swiss electricity consumption

Leveraging Data for Energy Transformation: How Data Science can support Swiss Energy Policies?

Contact us

Let’s talk Data Science

Do you need our services or expertise?
Contact us for your next Data Science project!