<![CDATA[Stunting remains a persistent public health challenge in Indonesia, affecting 21.6% of children under five according to the 2022 Indonesian Nutrition Status Survey (SSGI), while nutritional status classification at Posyandu Makmur 1 is still performed manually using the WHO z-score table, a process prone to human error. This study integrates a Random Forest algorithm, optimized through 5-fold Grid Search Cross-Validation, to classify toddler nutritional status into four categories (severe malnutrition, undernutrition, normal nutrition, and overnutrition) based on anthropometric data, and implements the resulting model into a Flask-based web application developed using the Rapid Application Development (RAD) method. The model was trained on 263 records of toddlers aged 0-59 months (80:20 train-test split) and evaluated on 53 test samples, achieving 83.02% accuracy with n_estimators=200 and max_depth=15. However, additional evaluation using macro F1-score (59.90%) and balanced accuracy (61.64%) reveals a substantial performance drop for minority classes, particularly undernutrition, due to class imbalance in the dataset. Feature importance analysis identified height (35.5%) and age (30.1%) as the most influential predictors, consistent with the height-for-age Z-score formula used for data labeling. The resulting system passed both White Box and Black Box testing, confirming its functional readiness as an early screening tool at primary health services, and underscoring the need for class imbalance mitigation strategies in future machine learning-based nutritional classification research]]>
Copyrights © 2026
