Building an Intelligent System for Food Distribution Empowered by AI: Tackling Surplus–Deficit Inequality in the Barlingmascakeb Agglomeration

Abstract

Food distribution disparities remain a persistent challenge in the Barlingmascakeb region (Banyumas, Cilacap, Purbalingga, Banjarnegara, and Kebumen), where socio-economic and infrastructural factors drive regional inequalities. This study applies a machine learning–based classification approach to identify sub-districts categorised as food surplus or deficit. The dataset, initially imbalanced, was balanced using the Synthetic Minority Oversampling Technique (SMOTE), followed by training and evaluating four ensemble algorithms: AdaBoost, Gradient Boosting, XGBoost, and CatBoost. Among the tested models, AdaBoost demonstrated the best overall performance with an accuracy of 0.9565, precision of 1.00, recall of 0.8333, and F1-score of 0.9091. Gradient Boosting achieved a more balanced recall (0.8333) than XGBoost and CatBoost, although with lower precision. Based on the Gradient Boosting model, Feature importance analysis identified the Food Security Index as the most critical determinant of food status, followed by clean water access, morbidity rate, health workforce availability, and poverty levels. This study offers a novel contribution by providing a high-resolution, sub-district-level classification of food surplus and deficit conditions using interpretable ensemble machine learning models integrated with multidimensional socio-economic and health indicators. Practically, the model supports targeted and data-driven food distribution policies; theoretically, it reinforces the multifaceted nature of food security beyond production alone; and for future research, it opens opportunities to extend the framework to spatio-temporal and optimization-based food distribution models.