Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The 2023/24 VIEWS Prediction Challenge: Predicting the Number of Fatalities in Armed Conflict, with Uncertainty

Håvard Hegre, Paola Vesco, Michael Colaresi, Jonas Vestby, Alexa Timlick, Noorain Syed Kazmi, Friederike Becker, Marco Binetti, Tobias Bodentien, Tobias Bohne, Patrick T. Brandt, Thomas Chadefaux, Simon Drauz, Christoph Dworschak, Vito D'Orazio, Cornelius Fritz, Hannah Frank, Kristian Skrede Gleditsch, Sonja Häffner, Martin Hofer, Finn L. Klebe, Luca Macis, Alexandra Malaga, Marius Mehrl, Nils W. Metternich, Daniel Mittermaier, David Muchlinski, Hannes Mueller, Christian Oswald, Paola Pisano, David Randahl, Christopher Rauh, Lotta Rüter, Thomas Schincariol, Benjamin Seimon, Elena Siletti, Marco Tagliapietra, Chandler Thornhill, Johan Vegelius, Julian Walterskirchen

TL;DR

This paper presents the 2023/24 VIEWS Prediction Challenge, which tasks teams with probabilistic forecasts of monthly armed-conflict fatalities, incorporating uncertainty through predictive samples and distributional outputs. It defines dual forecasting units—country-months ($cm$) and PRIO-GRID months ($pgm$)—across two horizons: the true future (July 2024–June 2025) and historical windows (2018–2023), with forecasts submitted as samples or Poisson-based point predictions. The evaluation framework centers on the Continuous Rank Probability Score ($CRPS$) as the main metric, complemented by the log score and Mean Interval Score ($MIS$), emphasizing calibration, sharpness, focus, nearness, and propriety, and it includes a suite of official benchmark models and an ensemble procedure. The challenge aims to advance probabilistic forecasting in conflict fatalities, encourage methodological innovation, and facilitate reproducibility, with a dedicated visualization tool and a follow-up evaluation article planned for Fall 2025. The practical impact lies in providing decision-makers with calibrated predictive distributions of fatalities to inform risk assessment and early-warning actions in conflict settings.

Abstract

This draft article outlines a prediction challenge where the target is to forecast the number of fatalities in armed conflicts, in the form of the UCDP `best' estimates, aggregated to the VIEWS units of analysis. It presents the format of the contributions, the evaluation metric, and the procedures, and a brief summary of the contributions. The article serves a function analogous to a pre-analysis plan: a statement of the forecasting models made publicly available before the true future prediction window commences. More information on the challenge, and all data referred to in this document, can be found at https://viewsforecasting.org/research/prediction-challenge-2023.

The 2023/24 VIEWS Prediction Challenge: Predicting the Number of Fatalities in Armed Conflict, with Uncertainty

TL;DR

This paper presents the 2023/24 VIEWS Prediction Challenge, which tasks teams with probabilistic forecasts of monthly armed-conflict fatalities, incorporating uncertainty through predictive samples and distributional outputs. It defines dual forecasting units—country-months () and PRIO-GRID months ()—across two horizons: the true future (July 2024–June 2025) and historical windows (2018–2023), with forecasts submitted as samples or Poisson-based point predictions. The evaluation framework centers on the Continuous Rank Probability Score () as the main metric, complemented by the log score and Mean Interval Score (), emphasizing calibration, sharpness, focus, nearness, and propriety, and it includes a suite of official benchmark models and an ensemble procedure. The challenge aims to advance probabilistic forecasting in conflict fatalities, encourage methodological innovation, and facilitate reproducibility, with a dedicated visualization tool and a follow-up evaluation article planned for Fall 2025. The practical impact lies in providing decision-makers with calibrated predictive distributions of fatalities to inform risk assessment and early-warning actions in conflict settings.

Abstract

This draft article outlines a prediction challenge where the target is to forecast the number of fatalities in armed conflicts, in the form of the UCDP `best' estimates, aggregated to the VIEWS units of analysis. It presents the format of the contributions, the evaluation metric, and the procedures, and a brief summary of the contributions. The article serves a function analogous to a pre-analysis plan: a statement of the forecasting models made publicly available before the true future prediction window commences. More information on the challenge, and all data referred to in this document, can be found at https://viewsforecasting.org/research/prediction-challenge-2023.
Paper Structure (48 sections, 4 equations, 4 figures, 3 tables)

This paper contains 48 sections, 4 equations, 4 figures, 3 tables.

Table of Contents

  1. Motivation
  2. The challenge
  3. Prediction target: Predicting the monthly number of fatalities from organized political violence
  4. 1+6 prediction windows
  5. State-based conflict
  6. The VIEWS levels of analysis
  7. Country-months
  8. PRIO-GRID months
  9. Timeline and format of contributions
  10. The contributions
  11. Drauz_Becker_quantile_forecast
  12. Bodentien_Rueter_NB
  13. CCEW_tft
  14. CCEW_trees_global
  15. ConflictForecast
  16. ...and 33 more sections

Figures (4)

  • Figure 1: Distribution of (logged) observed fatalities in December 2022 at our two levels of analysis, cm (left) and pgm (right), and for different types of violence reported by the UCDP-GED Dataset Davies2022JPR: state-based ("ln_ged_sb", blue), non-state ("ln_ged_ns", orange) and one-sided ("ln_ged_os", green).
  • Figure 2: ViEWS ensemble point predictions for January 2024, cm and pgm level
  • Figure 3: Predicted fatalities in December 2022 at our two levels of analysis
  • Figure 4: Illustration of the contribution to the CRPS metric for one instance from the observed data, based on Bracher2021PLOS