Regional Clustering of Food Insecurity to Support the Attainment of SDG 2: Zero Hunger through Machine Learning Approaches

Abstract

Food security remains a persistent development challenge in Indonesia, with regional disparities posing significant barriers to achieving equitable access to nutritious and sufficient food. This study aims to classify and cluster districts and cities in Indonesia based on their food security vulnerability levels, thereby supporting the attainment of SDG 2: Zero Hunger. We employed a machine learning approach using a dataset of 514 regions and nine food security indicators sourced from national databases. The classification phase compared three algorithms, Random Forest, XGBoost, and LightGBM, under multiple data preprocessing scenarios, including outlier handling (IQR and Isolation Forest) and class balancing (SMOTE). LightGBM with IQR preprocessing delivered the best performance, achieving an accuracy and F1-score of 0.984. For clustering, DBSCAN and HDBSCAN were applied using the six most important features identified by the classifier. DBSCAN showed slightly better performance based on Silhouette Score (0.5639), resulting in three regional groupings: food-secure, highly vulnerable, and outlier regions. The analysis revealed that socio-economic factors and access to basic infrastructure remain critical determinants of food insecurity. The results underscore the importance of data-driven approaches in policy formulation and highlight the value of machine learning in producing more targeted, efficient, and adaptive food security interventions in Indonesia.