The final submissions to the VIEWS Prediction Challenge 2023/2024 have trickled in and we are thrilled to announce the release of the conflict forecasts from our participating teams, the launch of the VIEWS Prediction Challenge dashboard, and the preprint publications from the challenge.

Purpose of the challenge

Following recent calls for “Early Warning Early Action” (EWEA) amongst international governmental organisations (IGOs) such as the United Nations, and government ministries across the world, there is a rising demand for live early-warning systems (EWS) that can help predict incidence, intensity, and onset of armed conflict. It may support decision-makers’ preparation for, mitigation of, and even prevention of humanitarian crises.

Numerous systems, such as our own Violence & Impacts Early-Warning System (VIEWS), offer such services today but are limited to providing forecasts for the most likely outcome, often in the form of point predictions. These models perform well at predicting no violence and continuation of ongoing conflicts, yet by solely relying upon point predictions, their utility is less relevant for necessary policy actions related to outbreak of new conflict or sudden escalation of existing violence. Forecasts with probability distributions, on the other hand, can signal both what the most likely outcome will be and alert to any low but alarming future risks.

With the goal to advance research on predicting with uncertainty and ultimately seek to enhance the potential use of early warning for early action, the VIEWS team launched its second prediction challenge in 2023 (read more about the first prediction challenge in International Interactions 48(4)). Participating teams were invited to forecast fatalities as a probability distribution over the “best” estimate of fatalities in state-based armed conflict per the UCDP definitions, which therefore fully take uncertainty of forecasts into account.

The contributions

Following the deadline on June 23rd, the final submissions have now trickled in and we are thrilled to present the forecasts from the 13 participating teams in this year’s edition of the VIEWS Prediction Challenge.

No less than 23 models have been developed by the participants over the course of the challenge. Each model produces predictions for the number of fatalities in state-based armed conflict with uncertainty from July 2024 to June 2025, in addition to a test set for the full calendar years of 2018 to 2023.

The models offer an exciting mix of country-month forecasts with global coverage, and PRIO-GRID-month predictions for Africa and the Middle East. All predictions have been collated into percentiles by the VIEWS team and are available for exploration on the brand new VIEWS Prediction Challenge Dashboard! It was specifically developed to display this year’s challenge results.

Launch of the VIEWS Prediction Challenge dashboard

The Prediction Challenge dashboard was launched on July 1st with the percentile forecasts from all participating teams and five VIEWS benchmark models. It allows users to navigate between mapped forecasts from all models, available levels of analysis, percentiles, years and months of the forecasts, which is further complemented by dynamic 12-month line charts of the predictions.

From August 2024 onwards, the dashboard will be expanded with rolling evaluations of the participants’ predictions, assessed against the monthly releases of recorded conflict data from the UCDP Candidate Events dataset. It will also display evaluations of the forecasts for the historic test set, 2018-2023, based on the annual UCDP GED datasets.

Introduction article and model papers

Learn more about the prediction challenge in the joint introduction article (preprint) and the model documentation in working papers from the participating teams. These articles provide in-depth information on the procedures, evaluation metrics, and most importantly the unique contributions of each team to the field of forecasting. Fundamentally, the articles are public statements of the forecasting models prior to the true future prediction window commencing.

To receive monthly updates on the prediction challenge, please subscribe to the VIEWS newsletter.

Acknowledgements

The Prediction Challenge 2023/2024 was made possible with funding from the German Ministry of Foreign Affairs and PREVIEW.

We would also like to thank the challenge hosts and organisers Paola Vesco, Håvard Hegre, Michael Colaresi, and Jonas Vestby for their hard work facilitating this challenge; the many members of the VIEWS team who have supported and contributed to the project; our consultant Henrik Valerian who built and customised the VIEWS Prediction Challenge Dashboard to share the results from the project; and – of course – our participating teams!