Predicting Epidemiological and Sociopolitical Agrifood Shocks: An Early Warning Index for Africa

Abstract

African agrifood systems provide livelihoods to about 1.6 billion people and employ close to 60% of Africa's population. Existing frameworks for measuring agrifood resilience focus primarily on climatic and economic shocks largely overlooking epidemiological outbreaks and armed conflict as distinct drivers of agrifood disruption. Across the forty African countries examined in this study, disease outbreaks can increase mortality, restrict population mobility, reduce labour availability, disrupt agricultural production and markets, and weaken household purchasing power. For example, during the 2014–2016 West African Ebola epidemic, rice production dropped by an estimated 12% in Liberia, 8% in Sierra Leone, and 4% in Guinea. These disruptions exacerbated food insecurity,and rural poverty, underscoring how vulnerable agrifood systems are to epidemiological shocks.

In this work, we examine how disease outbreaks and armed conflict propagate through agrifood systems by disrupting labour availability, agricultural production, food prices, markets, and household food security. To capture these dynamics, we developed a multidimensional vulnerability framework based on 51 indicators across four categories: Shock Exposure, Agrifood Stress, Labour & Youth, and Low Response Capacity. The framework integrates region-level difference-in-differences models, GIS overlay analysis, and agent-based modelling to construct a 0–100 Agrifood Vulnerability Index across the forty African countries. Utilizing historical crop production, food-price, disease, and conflict data for selected African regions, we trained a machine-learning model to identify conditions associated with escalating agrifood disruption. Following multicollinearity screening using correlation analysis and variance inflation factors, the best-performing Random Forest model classified high-risk agrifood-disruption periods with 92% accuracy, a receiver operating characteristic area under the curve of 0.94, and an F1 score of 0.90 on the held-out test set. Performance remained consistent across five-fold cross-validation, with a mean accuracy of 89%. Shock intensity, food-price volatility, conflict-related displacement, agricultural-production decline, and low response capacity emerged as the strongest predictors of elevated vulnerability. The model identified high-risk conditions approximately three months before observed deterioration in food-security outcomes. Ultimately, this framework serves as an early warning system to guide targeted intervention that safeguard vulnerable populations, agricultural livelihoods, and broader agrifood systems by extending resilience assessment beyond climate and economic shocks to include epidemiological and sociopolitical risks.