The fatalities003 model

Conflict prediction model in use from November 2025 onwards
Important notice: In November 2025, we made new forecasts available from both the  fatalities003 and fatalities002  models to allow for comparison. Both datasets are available through the VIEWS API as well as for download in the data section of this website. The data dashboard is however still limited to forecasts from the fatalities002 model. From December 2025 onwards, the fatalities002 model will be discontinued and replaced by fatalities003.

Model OVERVIEW

Scope and coverage

Source code
The model generates monthly predictions for the number of fatalities in impending state-based armed conflict, as well as dichotomous forecasts for the probability of at least 25 battle-related deaths (BRDs) per country-month and at least 1 BRD per PRIO-GRID-month. 
  • Hover over the boxes for more information.

UPDATE SCHEDULE

Monthly

PREDICTED TYPE(S) OF VIOLENCE

State-based conflict

FORECASTING WINDOW

1-36 months ahead

PREDICTED OUTCOMES

Continuous & dichotomous predictions

COUNTRY-LEVEL COVERAGE

Global

SUB-NATIONAL COVERAGE

Africa and the Middle East

INPUT DATA

The indicators informing the fatalities003 model

The fatalities003 model is informed by nearly 200 variables from data providers such as the Uppsala Conflict Data Program (UCDP), ACLED, PRIO-GRID, the World Bank, V-Dem, and FAO.

MODEL DETAILS

Delve deeper into the model specifications

The fatalities003 model is specified in an open-source code repository on GitHub: views-models. To consult the production model specifications, navigate to the ensembles called pink_ponyclub (the country-level ensemble) and skinny_love (the PRIO-GRID level ensemble). The November 2025 forecasts were produced from the development branch of the repository. 

DIFFERENCES

How the model differs from its predecessor

Note that the list of source indicators informing the fatalities003 model, as well as the means in which the sub-models are implemented in the latter, differs from the predecessor model (fatalities002). The ensembling technique at the country level has also changed from a genetic algorithm to a simple unweighted average. Last, the surrogate model forecasts available from the  fatalities002 model have now been discontinued. The main forecasts from the models, however, remain the same: both models produce forecasts for the number of fatalities from state-based armed conflict, in both logged and non-logged form. They both also produce dichotomous forecasts for the probability of observing at least 25 battle-related deaths per country-month and at least one battle-related death per PRIO-GRID cell and month.