Predicting Household Food Insecurity Through a Livelihood lens: A Machine Learning Approach in Rural Colombia

ABSTRACT Food insecurity remains a persistent challenge in rural Colombia, driven by structural inequalities, livelihood constraints and exposure to shocks. This paper examined the determinants and predictive drivers of household food insecurity using nationally representative data from the Food and Agriculture Organisation (FAO) Data in Emergencies Monitoring (DIEM) surveys collected in 2023, covering 2771 rural households. The analysis integrated machine learning techniques with logistic regression to capture both predictive performance and interpretability, guided by the Sustainable Livelihoods Framework. The results showed that ensemble machine learning models achieved strong predictive accuracy, with the up‐sampled logistic regression model performing competitively (AUC = 0.758), confirming its suitability as a benchmark. Explainable machine learning and regression estimates consistently identified education, income, household size, agricultural participation and exposure to shocks as key determinants of food insecurity. In particular, participation in agriculture reduced the likelihood of food insecurity, while higher food prices, health‐related shocks and reliance on adverse coping strategies significantly increased vulnerability. Geographic disparities were also evident, with households in regions such as Bolívar, Cesar and La Guajira facing substantially higher risks. The findings demonstrated that food insecurity reflects the interaction of structural, livelihood and shock‐related factors rather than isolated effects. Integrating predictive modelling with a livelihood‐based framework improved the identification of high‐risk households and strengthened the evidence base for targeted, timely and context‐specific policy interventions to reduce food insecurity in rural Colombia.