<![CDATA[Stunting remains a pressing health issue in Indonesia’s coastal communities, where uneven access to nutrition services and maternal–child care can limit early prevention efforts. This study develops a machine learning framework to estimate stunting risk among children under five in the eastern coastal region of Aceh. The modeling process began with data cleaning and preparation, followed by class rebalancing with SMOTE and automated parameter tuning using Optuna. Four standalone classifiers were evaluated: Logistic Regression, Gaussian Naïve Bayes, Support Vector Machine, and Random Forest. Their outputs were then extended through three ensemble configurations, namely XGBoost-assisted boosting for each baseline model, a stacking scheme named Stackstun with Logistic Regression as the final learner, and an optimized weighted soft-voting model referred to as Votsstun. Performance was measured using accuracy, precision, recall, and F1-score. Among the single classifiers, Logistic Regression achieved the best result, with an accuracy of about 0.93. The strongest overall performance was obtained by the boosted Random Forest model, which reached an accuracy of 0.9952 and produced almost perfect class-level precision, recall, and F1-score. Votsstun also performed consistently, recording an accuracy of approximately 0.986, while its macro and weighted F1-scores approached 0.99. These results indicate that the combined use of class rebalancing, automated optimization, and ensemble learning can improve the robustness of stunting-risk classification. The proposed framework may assist local health agencies in identifying vulnerable children earlier, prioritizing limited intervention resources, and strengthening prevention programs in coastal communities.]]>
Copyrights © 2026
