Known issues
Data-driven conflict forecasting models are vital tools in preventing violence and mitigating the impacts of armed conflicts on society and humanitarian outcomes. However, like any forecasting method, these models come with their own
set of limitations and may be affected by model bugs or other errors. It is crucial for academic researchers and practitioners to be aware of these in order to effectively utilize the models and interpret their results. On this page,
we document issues and limitations that are specific to the operational VIEWS model, from a user perspective.
Labels
Issue categories
Records on this page are labeled by type of issue – by the source of the issue(s) at hand. This allows separation between problems that stem from the model code itself (bugs); issues derived from input data that may have implications
on the model’s predictions; and general limitations that may be important for researchers and practitioners to keep in mind when using our data.
- Bug refers to model bugs (detected errors) in the model code.
- Input data refers to issues derived from the data that the model is trained on, such as temporary cases of high uncertainty in recorded fatalities (noted for example during the Tigray war in Ethiopia, during which recorded fatalities from the conflict deviated significantly from the reality on the ground for several months due to the media blackout), or technical issues related to ingestion of new data into the VIEWS system (caused by, e.g., broken APIs).
- Limitation refers to noted limitations of the model that stem from methodological choices or model design rather than bugs or errors in the production pipeline. This includes, e.g., choice of GIS dataset(s) to determine country borders, where used borders may differ from those applied by governments, IGOs, or NGOs; or definitions of armed conflict. While issues listed with this label are not errors per se, they reflect limitations may be important to keep in mind when using outputs from the model, or comparing its results to other datasets or expert assessments (which may apply other definitions).
Priority status
All issues listed on this page, and in the dedicated GitHub repository, have been noted and assessed by the VIEWS team. While bugs and issues related to input data have priority for resolution, also inherent limitations of the model
may be subject to revision. For transparency towards the users of VIEWS data, we assign a priority status to all open issues, ranging from high priority to tabled issues. The latter is reserved for issues that we are aware of and
seek to address in the future, but cannot allocate resources to at this point in time.
Open issues
Current issues related to the VIEWS model and output
Static overview of the Open issues recorded in the
VIEWS Outreach repository. For up-to-date information regarding each issue, please consult the repository directly.
All clear! There are no recorded issues at this time.
This page lists open issues from a user perspective; issues or limitations of particular importance for researchers and practitioners using our data across the triple nexus of peace, development, and humanitarian assistance.
Developers seeking to peek into our source code and model specifications are kindly directed to the open-source repository
viewsforecasting.
Closed issues
Resolved issues
Static overview of the Closed issues recorded in the VIEWS Outreach repository. For up to date information regarding each issue, please consult the repository directly.
Incomplete ingestion of ACLED data from Sept-Oct 2023
Ingestion of ACLED data for September and October 2023 failed due to issues processing data from the provider’s API.
Identified bug related to handling missing data in the Markov models
Bug identified in the Markov models (constituent models specialised at predicting conflict onset) related to handling missing data. The bug caused the models to treat missing values in their training and calibration datasets
as errors rather than zeroes, consequently dropping countries with NaN data values therein.