Economics of Disasters and Climate Change Гуманитарные и социальные науки
Predictive Modeling of Household Food Insecurity Using Machine Learning: Evidence from Somaliland Household Budget Survey, 2022
Аннотация
Background Food insecurity remains a persistent challenge in Somaliland, where households are exposed to overlapping socio-economic, demographic, and climate-related stresses. Climate exposure was measured using household-reported drought, flood, and fire-related shocks. Previous empirical studies in East Africa and Somalia have mainly relied on traditional econometric approaches such as logistic regression, which assume linear relationships and are limited in capturing compound and nonlinear risk structures, particularly the interaction between climate shocks and household characteristics. This study addresses this methodological limitation by applying a comparative machine learning framework to predict household food insecurity. Methods The analysis uses survey-weighted household data from 2,112 observed households in the Somaliland Integrated Household Budget Survey (SIHBS) 2022. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied only to the training dataset after the train–test split. Five models were implemented and compared: Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, and Neural Network. Model performance was evaluated using accuracy, balanced accuracy, and ROC-AUC. Results The findings show that nonlinear and ensemble models outperform the linear approach, with Random Forest achieving the highest predictive performance (ROC-AUC = 0.915; accuracy = 0.862), followed by Decision Tree (ROC-AUC = 0.831) and Neural Network (ROC-AUC = 0.818). Feature importance and SHAP analyses consistently identify household size, regional location, and climate-related shocks as the most influential predictors. Conclusions The study concludes that machine learning, particularly ensemble methods, provides an effective framework for identifying vulnerable households and supporting targeted food-security interventions.