Forecasting change in conflict fatalities with dynamic elastic net

This article illustrates an approach to forecasting change in conflict fatalities designed address the complexity of drivers and processes armed conflicts. The design this is based on two main choices. First, account for specificity over time space, we model conflicts each individual country separately. Second, draw adaptive model—Dynamic Elastic Net, DynENet—which able efficiently select relevant predictors among a large set covariates. We include 700 variables our models, adding event data top features provided by convenors competition. show that suitable computationally efficient enough dynamics. Moreover, nature brings significant added value. Because only selects are predict intensity, retained can be analyzed describe dynamic configuration both across countries within time. Countries then clustered observe emergence broader patterns related correlates conflict. In sense, produces interpretable forecasts, addressing one key limitation contemporary approaches forecasting.