Mass atrocities bring untold devastation, uprooting of families, and can leave hundreds, even thousands of humans in extremely dire circumstances. A better understanding of the indicators of vulnerability – conditions that lead to atrocity – are helping to better model, and therefore predict where a mass atrocity is most likely to occur. Through better data solutions, prediction can then lead to prevention.
For this challenge, the client was seeking an improved, statistical model to out-predict existing subjective analysis models. However, there was no proof that an answer was present in the available data sources.
The challenge was structured in a series to first identify & validate existing and new public data sets on national and sub-national violence.
Then, a predictive modeling challenge was launched, using newly validated data sets.
The challenge resulted in 5 winning and powerful new data solutions that can be easily used for future improvement.
The newly created predictive models help identify community-level risk factors that make communities more or less likely to experience acts of mass violence.
The top solutions demonstrate a significant event-prediction correlation, meaning the core assumption that an improved statistical model could be created, was proven true.