#Forecasting the Relative Risk for the Onset of Mass Killings to Help Prevent Future Atrocities
##Analytic Challenge
Mass atrocities are rare yet devastating crimes. They are also preventable. Studies of past atrocities show that we can detect early warning signs of atrocities and that if policy makers act on those warnings and develop preventive strategies, we can save lives. Yet despite this awareness, all too often we see warning signs missed and action taken too late, if at all, in response to threats of mass atrocities.
The Early Warning Project, an initiative of the United States Holocaust Memorial Museum (Holocaust Museum), aims to assess a country’s level of risk for the onset of future mass killings. Over time, the hope is to learn which models and which indicators are the best at helping anticipate future atrocities to aid in the design and implementation of more targeted and effective preventive strategies. By seeking to understand why and how each countries’ relative level of risk rises and falls over time, the system will deepen understanding of where new policies and resources can help make a difference in averting atrocities and what strategies are most effective. This will arm governments, advocacy groups, and at-risks societies with earlier and more reliable warning, and thus more opportunity to take action, well before mass killings occur.
The project’s statistical risk assessment seeks to build statistical and machine learning algorithms to predict the onset of a mass killing in the succeeding 12 months for each country with a population larger than 500,000. The publically available system aggregates and provides access to open source datasets as well as democratizes the source code for analytic approaches developed by the Holocaust Museum staff and consultants, the research community, and the general public. The Holocaust Museum engaged Booz Allen to validate existing approaches as well as explore new and innovative approaches for the statistical risk assessment.
##Our Approach
Taking into account the power of crowdsourcing, Booz Allen put out a call to employees to participate in a hack-athon— just the start of the team’s support as the Museum refined and implemented the recommendations. More than 80 Booz Allen Hamilton software engineers, data analysts, and social scientists devoted a Saturday to participate. Interdisciplinary teams spent 12 hours identifying new datasets, building new machine learning models, and creating frameworks for ensemble modeling and interactive results visualization. Following the hack-a-thon, Booz Allen Data Scientists worked with Holocaust Museum staff to create a data management framework to automate the download, aggregation, and transformation of the open source datasets used by the statistical assessment. This extensible framework allows integration of new datasets with minimal effort, thereby supporting greater engagement by the Data Science community.
##Our Impact
Publically launched in the fall of 2015, the Early Warning Project can now leverage advanced quantitative and qualitative analyses to provide governments, advocacy groups and at-risk societies with assessments regarding the potential for mass atrocities around the world. Integration of the project’s statistical risk assessment models and expert opinion pool created a publicly available source of invaluable information and positioned Data Science at the center of global diplomacy.
The machine learning models developed during the hack-a-thon achieved performance on par with state of the art approaches as well as demonstrated the efficacy of predictions 2-5 years into the future. Teams also identified approaches for constructing test/validation sets that support more robust model evaluation. These risk assessments are an important technological achievement in and of themselves, but what this initiative means for the Data Science community’s position in global diplomatic dialogue marks an entirely new era for those on the frontiers of Big Data.
The data management framework developed from the lessons learned of the hacka- thon represents a great leap forward for the Holocaust Museum. The periodicity of aggregating and transforming data was reduced from twice per year to once per week. In addition to providing the community with more up-to-date data, the reduced burden on researchers enables them to spend more time analyzing data and identifying new and emergent trends. The extensible framework will also allow the Holocaust Museum to seamlessly integrate new datasets as they become available or are identified by the community as holding analytic value for the problem at hand.
Through this project, the Holocaust Museum was able to shift the dynamic from monitoring ongoing violence to determining where it is likely to occur 12 to 24 months into the future by integrating advanced quantitative and qualitative analyses to assess the potential for mass atrocities around the world. The Early Warning Project is an invaluable predictive resource supporting the global diplomatic dialogue. While the focus of this effort was on the machine learning and data management technologies behind the initiative, it demonstrates the growing role the Data Science community is playing at the center of global diplomatic discussions.