Integrated Machine Learning Framework for Smart Food Systems Optimization in Aceh, Indonesia

Abstract

Effective management of smart food systems in Aceh, Indonesia, requires accurate forecasting, assessment of regional vulnerability, and robust supply chain monitoring. This study presents a multistage machine learning (ML) framework integrating predictive modeling, clustering optimization, and classification to support evidence-based food management decisions. Ridge regression projected food commodity prices for 2025–2028, achieving an average mean squared error (MSE) of 43.045.946 and an average root mean squared error (RMSE) of 4.352,238, capturing trends for both staple and high-value commodities. Regional food vulnerability is evaluated using K-means clustering enhanced with simple additive weighting (SAW)-based centroid initialization, which reduces the average number of iterations to 7.5 compared to 8.7 for standard K-means. The SAW-enhanced clustering achieves an average Calinski–Harabasz score of 40.887 and an average Silhouette Score of 0.288, generating three coherent clusters: Food Secure, Food Vulnerable, and Food Insecure. Classification of regional supply chain stability using support vector machines (SVM) with radial basis function (RBF) and sigmoid kernels alongside random forest (RF) demonstrates that SVM-RBF attains 94.59% accuracy, SVM-sigmoid reaches 46%, and RF achieves a mean 10-fold cross-validation accuracy of 98.89% with low variability and F1-scores ranging from 0.985 to 0.995. By integrating predictive, clustering, and classification analyses, this framework provides actionable insights that enable policymakers to anticipate price fluctuations, identify vulnerable districts, and implement targeted interventions, thereby enhancing food security and resilience in Aceh. The proposed methodology highlights the value of combining multiple ML approaches for optimizing smart food systems in regional contexts.