Predicted number of fatalities from state-based violence in October 2022, based on input data up to and including July 2022.
The Violence Early-Warning System (ViEWS) is a publicly available data-driven forecasting system at the frontier of research that generates monthly predictions of conflict fatalities up to 36 months ahead – throughout Africa and the Middle East. The project launched in 2017 to help policy-makers and practitioners plan anticipatory action and humanitarian interventions with a transparent and evidence-based approach. It is based at Uppsala University and Peace Research Institute Oslo.
Wars are highly destructive – beyond the immediate mortality and morbidity, conflicts lead to profound long-term consequences that impede the realization of sustainable development. They severely hinder growth and poverty reduction, typically reducing per-capita growth rates at least 2% annually. They also have severe detrimental effects on public health, across a wide variety of causes of death, with the adverse impact felt for decades after the shooting stops – conflict is ‘development in reverse’.
Accordingly, preventing and containing armed conflict is high on policy-makers’ agenda, but early action requires early warning. With a good understanding of where armed conflict will occur in the near future, for how long it will persist, and how lethal it will be, concerned governments and international governmental organisations (IGOs) may engage in diplomacy efforts, alert the international community about the situation, prepare for humanitarian assistance, and allocate resources where most needed.
With funding from a European Research Council Advanced Grant, the Violence Early-Warning System (ViEWS) was launched as a five-year research project at Uppsala University back in 2017, set out to build a pilot conflict forecasting system for the African continent. Serving as a successful proof of concept of how to meet the need above and bridge the knowledge gaps concerning areas of risk, the project has since grown exponentially by means of continous development, testing, and iterative improvements with additional funding from e.g. UN ESCWA, UNHCR, and UK FCDO.
Today, ViEWS is a live forecasting system based at both Uppsala University and Peace Research Institute Oslo with a core team of about 10-15 people and research associates in the dozens. It systematically and uniformly monitors and assesses the risks of conflict at a high geographic resolution, generating monthly predictions for the expected number of conflict fatalities during each month in a rolling three-year window – for each country across the globe as well as for sub-national (0.5x0.5 decimal degree; approximately 55 km at the Equator) locations across Africa and the Middle East.
The ViEWS system is built using state-of-the-art prediction methodologies and a strong body of empirical research on the drivers of conflict. It is constructed as groups of forecasting models (ensembles) that help trace the predictions back to the driving features, also accounting for the relative importance of different sets of conflict drivers when forecasting a few months relative to several years into the future. Overall, the current iteration of the system is trained on 535 million data points from 30+ years of time-series data on more than 250 different conflict-related variables, woven together into a theoretically and methodologically consistent data-driven prediction model. Together, these features make ViEWS one of – if not the – most sophisticated open-source systems of its kind.
Next on the project’s agenda is to expand to global scope at both levels of analysis, and to move from only predicting conflict to also predicting humanitarian impacts of armed conflict – transforming the ‘Violence Early-Warning System (ViEWS)’ to the ‘Violence & Impacts Early-Warning System (VIEWS)’. The impacts research will be conducted over the course of two new (separate but interlinked) six-year research projects under the leadership of ViEWS' Director, Håvard Hegre, namely 'Societies at Risk' at Uppsala University, and 'ANTICIPATE' at Peace Research Institute Oslo.
What Makes Your Project Innovative?
Early-warning systems can greatly benefit all levels of society through the potential to assist policy-makers and practitioners in eliminating blind spots, justify targeted action, provide an evidence-based means to allocate scarce resources, and serve as scenario analysis tools for policy-making.
ViEWS surpasses its competitors, as it:
- Is the only system to produce forecasts for three types of violence at two levels of analysis;
- Uses the most comprehensive forecasting horizon (1-36 months) and set of input variables;
- Is highly accurate, typically predicting correctly 95% of all country-months in conflict with only 35% false positives;
- Has developed a procedure to evaluate forecasts for all steps (months) ahead against all months in the evaluation data partition, offering much more precise evaluations of performance than its competitors;
- Is highly cost-effective, being built on an infrastructure and automatization routine that allows for monthly updates with limited effort.
What is the current status of your innovation?
The ViEWS pilot (2017-22) achieved scientific breakthrough, which generated great interest from the international community and allowed us to make invaluable connections and secure funding for further developments in direct collaboration with our key user group.
We now generate global predictions at the country level, and cover both Africa and the Middle East sub-nationally. We also offer not only probabilistic assessments of the likelihood that a given conflict threshold will be surpassed in a certain time and place (the pilot), but also estimates of the number of fatalities expected in said conflict. The latter allowed us to embark on two new research projects through which we move from predicting conflict to also predicting – and understanding – the impacts of armed conflict on human development. Simultaneously, we are working closely with our key users to tailor the output from ViEWS to better meet the needs of each organisation and realize the full potential of our innovation.
Collaborations & Partnerships
Beyond the ERC and our host institutions, our key partners have been:
- UN ESCWA, with whom we expanded the geographic scope of the model, incorporated factors of particular importance to their operations, and developed an API through which users can explore our complete datasets, and/or easily feed them into internal risk dashboards for further analysis.
- UK FCDO, with whom we moved to predicting the number of fatalities in impending conflict, which opened up for research on conflict impacts.
Users, Stakeholders & Beneficiaries
The users of the ViEWS system are the key beneficiaries thereof. They range from students and researchers who use our open-source code and data for academic advancements, to policy-makers and practitioners who use the conflict predictions as a source of evidence-based and data-driven intelligence to assist and support dire policy decisions and strategic planning for targeted action or humanitarian interventions.
Results, Outcomes & Impacts
Using procedures subjected to careful academic peer review, we have carried out industry-standard out-of-sample evaluations of predictive performance with excellent results (typically predicting correctly 95% of all country-months in conflict with only 35% false positives), which have been presented in prestigious academic journals. This scientific breakthrough has generated significant interest in international fora, leading to a number of collaborations with governments, IGO agencies, and other research institutes, allowing us to test the applicability our research in different contexts and develop tailor-made solutions to maximise the usefulness of our research to a larger audience.
Our forecasts are now used by fellow researchers, governments, IGOs and NGOs alike, and we continue to closely engage with our key user groups to iteratively improve our forecasting system and provide a high-quality public good as we transition to also predicting impacts of armed conflict.
Challenges and Failures
One of the greatest challenges the project has faced concerns the development of a robust, flexible, and cost-efficient data infrastructure. Over the course of the project, we have been forced to conduct no less than two comprehensive re-writes of this system. We have only just now been able to launch an infrastructure that is sufficiently robust and flexible to support the maintenance of our complex forecasting model.
Relatedly, implementing these re-writes while simultaneously conducting research on how best to develop the forecasting model itself has been challenging at best, not to mention obtaining the hardware to run ViEWS in a university setting.
Last, securing funding for regular maintenance of ViEWS has been a standing challenge. Most funders seek to support new research or novel developments rather than the daily operations that are key to any successful project, making it difficult to remain an attractive employer and keep key staff for longer periods of time.
Conditions for Success
One of the keys to the success of the ViEWS project lies in its embedment in two of the leading conflict research institutions – the Department of Peace and Conflict Research at Uppsala University, and Peace Research Institute Oslo. The former has also included a close collaboration with the Uppsala Conflict Data Program (UCDP), a global leader in conflict data collection. Together, this has allowed us to develop a system of high quality with strong links to state-of-the-art research.
Moreover, the value of human and financial resources cannot be understated. In addition to the need for strong leadership, a dedicated team is crucial to remain motivated and productive in difficult times and unexpected set-backs. Investing in your staff is certainly a key to success.
Last, having resources that are ear-marked for outreach activities have been key to ViEWS' success, allowing us to remain tuned to developments in international fora and to generate interest amongst key user groups.
The ViEWS system holds great potential for replication to other regions – by the project team or external users. This has already been cemented by our expansion from Africa to the Middle East at the sub-national level, and globally on the country level. The only constraint to geographic scalability is data availability, most importantly high-quality conflict data. Monthly updates are currently available for all countries in world, facilitating our global scope at this level of analysis, but is limited to Africa and the Middle East at the geographically disaggregated level.
The prediction system is also highly suitable for replication to other forecasting tasks, such as moving from conflict prediction to also predicting impacts of armed conflict on human development. The project team is currently embarking on this task by means of two separate but interlinked research projects: 'Societies at Risk' at Uppsala University, and 'ANTICIPATE' at Peace Research Institute Oslo.
Most importantly, don't underestimate the resources required for software development, nor the importance of thorough documentation thereof. The latter applies to all key tasks and routines, which otherwise risk being stalled for long periods of time in case of high workloads, sudden illness, or leaves of absence.
Moreover, consider carefully how to best evaluate the performance of your systems, which in turn are instrumental to optimizing it. Such evaluation criteria must be developed in dialogue with users that have the best qualitative understanding of what systems should seek to achieve.
As of 2022, ViEWS generates monthly predictions for the number of fatalities in impending conflict, as well as probabilistic assessments of the risk that a given conflict threshold will be surpassed in certain month and location. Country-level predictions are offered on a global scale, while sub-national forecasts cover Africa and the Middle East.
We also release monthly results from a set of interpretation models that highlight the relative importance of selected input variables (such as conflict history, democracy indices, infant mortality rate, or population size) to facilitate a greater understanding of the key drivers behind our forecasts.
Users can explore and download the complete datasets via the ViEWS API, or feed them into risk dashboards for further analysis. Users can also explore the source code via open-source repositories on GitHub.
To learn more about ViEWS, please contact us at [email protected]
- Developing Proposals - turning ideas into business cases that can be assessed and acted on
- Implementation - making the innovation happen
- Evaluation - understanding whether the innovative initiative has delivered what was needed
23 January 2023