Data-driven conflict classification exposes weak predictive indicators

Abstract Models and theories of armed conflict are effective when tailored to distinct conflict types, but existing classifications are often heuristic. We introduce a data-driven classification that is empirically grounded, reproducible and consistent across multiple scales. We leverage fine-grained conflict data, which we map to climate, geography, infrastructure, economics, raw demographics and demographic composition in Africa. Using an unsupervised learning model, we identify three overarching conflict types: ‘major-unrest’ at densely populated, riparian regions with well-developed infrastructure; ‘local-conflict’ in moderately populated, socio-economically diverse regions and often confined within country borders; and ‘sporadic-spillover events’ in low-population, underdeveloped areas. The three types stratify into a hierarchy of factors that highlights population, infrastructure, economics and geography, respectively, as the most discriminative indicators. Specifying conflict-type negatively affects the predictability of conflict intensity such as fatalities, conflict duration and other measures of conflict size. The competitive effect is a general consequence of weak statistical dependence. Hence, the empirical and bottom-up approach reveals how armed conflicts stratify into three archetypes, yet cautions us about the inclusion of commonly used indicators into predictive modelling.