ABSTRACT Sudan has been racked by complex and volatile conflicts for decades, with a new round of civil conflict erupting in April 2023, leading to significant loss of life and property. Despite the ongoing violence, the associations between conflicts and local contextual factors remain ambiguous. In this study, we developed an event-grid dataset, consisting of 50,033 observations with 20 variables derived from nighttime light (NTL) data, OpenStreetMap data, and other geographic data. Machine learning algorithms were employed to model the outbreak of conflict in 2023. Furthermore, the SHapley Additive exPlanations method was utilized to explain the relationships between diverse explanatory variables and conflicts. The results indicate that eXtreme Gradient Boosting model outperforms other models, such as Categorical Boosting and Light Gradient Boosting Machine. Multiple factors jointly contribute to conflicts. NTL-derived variables and transportation-related variables emerge as the most influential factors, followed by climate and agricultural factors. Regions characterized by economic inequality and proximity to transportation hubs are found to be more prone to conflicts. Additionally, variables impact the outbreak of conflict not only individually but also through mutual interactions. Notably, this study enhances a comprehensive and quantitative understanding of conflicts in Sudan, providing valuable insights to support humanitarian aid efforts.
Read full abstract