Predicting the impact of armed conflict on vulnerability: a Machine Learning approach

Abstract:

Armed conflicts have been associated with a variety of detrimental impacts on human security and development, and represent a crucial vector of societal vulnerability to subsequent climate hazards. The burgeoning literature on climate security has highlighted that climate variability and natural disasters may indirectly increase conflict risk in vulnerable locations. However, scientifically sound knowledge of the impacts of armed conflicts on socio-economic vulnerability remains sparse, and more research is needed to understand the complex linkages between natural disasters, armed conflict, and societal vulnerability. This study fills the gap by empirically investigating the impacts of armed conflicts and natural disasters on subsequent levels of societal vulnerability to climate hazards. The paper uses global, time-varying data for 189 countries between 1995 and 2019, combining information on natural disasters, armed conflict, and socio-economic vulnerability. We apply a leave-the-future-out cross validation and an extreme gradient boosting algorithm to test the out-of-sample performance of armed conflict, alone or in combination with natural disasters, as a predictor of vulnerability. This machine learning approach enables us to overcome some of the empirical challenges that traditional statistical methods relying on reduced form regressions fail to solve.

Authors:

Mariagrazia D’Angeli and Paola Vesco