<![CDATA[Stunting remains a significant public health challenge, where interventions are often delayed as they occur postnatally. This research aims to shift the detection focus to the prenatal phase by comparing the performance of four Ensemble Learning algorithms: Random Forest, XGBoost, CatBoost, and Light Gradient Boosting Machine (LGBM). Using a maternal dataset from DPPKB Parepare City consisting of 871 respondents, the models were developed through a Stratified 5-Fold Group Cross Validation scheme to predict stunting risk based on clinical features of pregnant women. Experimental results show that LGBM is the most optimal algorithm, where the hyperparameter tuning process increased model performance to an F1-Score of 94.34% and an accuracy of 94.50%. Ablation analysis identified maternal age, height, and the age of the last child as the most dominant predictors. The best model was integrated into a web-based decision support system using cloud-based microservices architecture, featuring geospatial mapping. This study proves that the application of LGBM on prenatal maternal data can provide accurate early detection to support targeted nutritional interventions for Family Assistance Teams (TPK) in Parepare City, although its generalizability remains limited to local administrative characteristics and warrants further prospective external validation before broader deployment.]]>
Copyrights © 2026
