Periodic Reporting for period 4 - ViEWS (The Violence Early-Warning System: Building a Scientific Foundation for Conflict Forecasting)
Reporting period: 2021-07-01 to 2022-06-30
The aim of the ViEWS project was to build an ambitious political violence early-warning system as well as to research the fundamentals for such a system. We have developed a system that is useful in the sense of having uniform, global coverage and frequent updates to avoid blind spots, is transparent and publicly available to anyone so that all risk assessments may be traced back to a fully specified argument and publicly available data and tools. ViEWS builds on the most advanced theoretical and empirical research within the field, and is based on systematic evaluation of predictive performance on data not used to train the system.
Succeeding in this objective is important for society. Large-scale political violence kills and maims thousands of people every month across the globe. For every person killed, hundreds are forced to relocate within countries and across borders. Armed conflicts have disastrous economic consequences, undermine political systems and public services, prevent developing countries from escaping poverty, and hinder international actors in providing humanitarian assistance. The challenges of preventing, mitigating, and adapting to large-scale political violence are particularly daunting when it escalates in locations and at times where it is not expected. Policy-makers and first responders would benefit greatly from a system that systematically monitors all locations at risk of conflict and assesses the risks of conflict escalation. If fully successful, ViEWS will help crisis responders to prepare for and prevent conflict-induced humanitarian disasters. Such a system is useful for domestic actors and IGOs and ensures maximum transparency and credibility regarding decisions made on the basis of specific warnings. In addition, it provides the scientific benefits of better understanding the causes, connections and consequences of conflict.
The pilot early-warning system is publicly available at https://viewsforecasting.org/ and the source code at GitHub (see https://viewsforecasting.org/resources/#source-code).
Succeeding in this objective is important for society. Large-scale political violence kills and maims thousands of people every month across the globe. For every person killed, hundreds are forced to relocate within countries and across borders. Armed conflicts have disastrous economic consequences, undermine political systems and public services, prevent developing countries from escaping poverty, and hinder international actors in providing humanitarian assistance. The challenges of preventing, mitigating, and adapting to large-scale political violence are particularly daunting when it escalates in locations and at times where it is not expected. Policy-makers and first responders would benefit greatly from a system that systematically monitors all locations at risk of conflict and assesses the risks of conflict escalation. If fully successful, ViEWS will help crisis responders to prepare for and prevent conflict-induced humanitarian disasters. Such a system is useful for domestic actors and IGOs and ensures maximum transparency and credibility regarding decisions made on the basis of specific warnings. In addition, it provides the scientific benefits of better understanding the causes, connections and consequences of conflict.
The pilot early-warning system is publicly available at https://viewsforecasting.org/ and the source code at GitHub (see https://viewsforecasting.org/resources/#source-code).
ViEWS has developed the data-collection and programming routines required to run the pilot early-warning system. We developed procedures to publish monthly updated `candidate' conflict event data, correct errors, aggregate to our units of analysis, and make publicly available through an API. To sustain the production of regularly updated forecasts, ViEWS has evolved a dynamic information processing platform, a first of its kind in conflict forecasting. The platform identifies and implemented in code the best solutions available for all steps in the process: ingestion of new and updated input data, transformations of input data (e.g. spatial and temporal lags, missing data handling), model-level selection of features, time-shifting to allow us to predict for a wide range of forecasting horizons, statistical and machine-learning modeling, hyper-parameter tuning, model ensembling, calibration, storing of predictions and metadata, evaluation of performance, visualizations, and publication through an API. Most steps are documented in research papers and in detail on GitHub (https://viewsforecasting.org/resources/#source-code).
Another central task has to be formulate theoretically informed models to help building the forecasting system, as well as to use prediction methods to evaluate theoretical propositions.
Throughout the project, the team has addressed a number of methodological challenges at all stages of the forecasting process. In particular, we have concentrated on incorporating spatial and temporal dynamics across multiple levels of analysis to integrate, weight, and improve forecasts. The results from this research is also documented in the code as well as in research papers.
This infrastructure and the data it contains were used to develop the various versions of the pilot early-warning system as well as to support other sub-tasks of the project. As a further demonstration of the value of such an infrastructure was the arrangement of a prediction competition.
Another central task has to be formulate theoretically informed models to help building the forecasting system, as well as to use prediction methods to evaluate theoretical propositions.
Throughout the project, the team has addressed a number of methodological challenges at all stages of the forecasting process. In particular, we have concentrated on incorporating spatial and temporal dynamics across multiple levels of analysis to integrate, weight, and improve forecasts. The results from this research is also documented in the code as well as in research papers.
This infrastructure and the data it contains were used to develop the various versions of the pilot early-warning system as well as to support other sub-tasks of the project. As a further demonstration of the value of such an infrastructure was the arrangement of a prediction competition.
The most significant achievement is the establishment of ViEWS, a live early-warning system for armed conflict that is more comprehensive and performant than any other publicly available system. ViEWS is used regularly by various IGOs, embodies a wide range of methodological optimizations and is published through a series of monthly reports, an API, and a website (https://viewsforecasting.org).
Another significant achievement is a first-of-its-kind, efficient, robust, versatile, dynamic, and open-source computational infrastructure for producing regularly updated armed conflict forecasts, based on state-of-the-art machine-learning principles. The compilation and coordination of a broad set of best practices constitutes what can be called an entire new methodology for forecasting armed conflict. Most steps are documented in research papers and in the various GitHub repositories reported in https://viewsforecasting.org/resources/#source-code.
Another major achievement was the successful execution of a prediction competition, leveraging the infrastructure developed by the project. The competition sought contributions to the problem of forecasting changes in fatalities in armed conflict, what research by ourselves and others identified as one of the most difficult problems an armed conflict early-warning system must solve. The challenge was presented with a uniform dataset, a clearly defined problem, and evaluation metrics specified in advance. The competition attracted 15 teams, including several leading scholars within the field, and increased our understanding of useful predictors, optimal algorithms, and ensemble techniques. The contributors were from the US, Scandinavia, and a number of European countries, spanning the disciplines of statistics, computer science, political science, economics, and conflict research, and included some of the leading forecasting environments in our field. A major reason for the success was the ability to make accessible to all teams the computational infrastructure, associated standardized datasets, and a joint evaluation process conducted by team members.
A final achievement is the collaboration networks we have established with stakeholders. The PI has presented ViEWS to a long series of UN organizations and at multiple international early-warning workshops, arranged by UN agencies, the New York University Center for International Cooperation, as well as by the German and Dutch MFAs. Collaboration has been particularly deep with the UN Economic and Social Commission for West Asia (ESCWA). The UN ESCWA funded an extension of ViEWS to cover the Middle East and incorporate some new predictors, and publishes the product through a `dashboard' (https://risks.unescwa.org). ViEWS also obtained funding from the UNHCR to collaborate on a report on `predictive analytics' in the Sahel. Moreover, ViEWS has written reports for internal use by the UN ESCWA and other UN organizations, and the UK Foreign and Commonwealth Development Office (FCDO) (https://github.com/prio-data/FCDO_predicting_fatalities).
Another significant achievement is a first-of-its-kind, efficient, robust, versatile, dynamic, and open-source computational infrastructure for producing regularly updated armed conflict forecasts, based on state-of-the-art machine-learning principles. The compilation and coordination of a broad set of best practices constitutes what can be called an entire new methodology for forecasting armed conflict. Most steps are documented in research papers and in the various GitHub repositories reported in https://viewsforecasting.org/resources/#source-code.
Another major achievement was the successful execution of a prediction competition, leveraging the infrastructure developed by the project. The competition sought contributions to the problem of forecasting changes in fatalities in armed conflict, what research by ourselves and others identified as one of the most difficult problems an armed conflict early-warning system must solve. The challenge was presented with a uniform dataset, a clearly defined problem, and evaluation metrics specified in advance. The competition attracted 15 teams, including several leading scholars within the field, and increased our understanding of useful predictors, optimal algorithms, and ensemble techniques. The contributors were from the US, Scandinavia, and a number of European countries, spanning the disciplines of statistics, computer science, political science, economics, and conflict research, and included some of the leading forecasting environments in our field. A major reason for the success was the ability to make accessible to all teams the computational infrastructure, associated standardized datasets, and a joint evaluation process conducted by team members.
A final achievement is the collaboration networks we have established with stakeholders. The PI has presented ViEWS to a long series of UN organizations and at multiple international early-warning workshops, arranged by UN agencies, the New York University Center for International Cooperation, as well as by the German and Dutch MFAs. Collaboration has been particularly deep with the UN Economic and Social Commission for West Asia (ESCWA). The UN ESCWA funded an extension of ViEWS to cover the Middle East and incorporate some new predictors, and publishes the product through a `dashboard' (https://risks.unescwa.org). ViEWS also obtained funding from the UNHCR to collaborate on a report on `predictive analytics' in the Sahel. Moreover, ViEWS has written reports for internal use by the UN ESCWA and other UN organizations, and the UK Foreign and Commonwealth Development Office (FCDO) (https://github.com/prio-data/FCDO_predicting_fatalities).