Stunting Classification Analysis for Toddlers in Bojongsoang: A Data-Driven Approach

Stunting is one of the health problem priorities for children in Indonesia. Prevention of stunting in toddlers is needed to avoid the long-term effects for both the toddlers and the public. Stunting prevention can be done by monitoring the growth of toddlers. Therefore, a system that can predict stunting conditions in toddlers is needed. Machine learning offers many methods that can be used to build a system to predict stunting conditions in toddlers. This research analyzes some machine learning models that are potentially suitable to predict stunting classes, which are K-Nearest Neighbor (KNN), Random Forest (RF), and Ensemble Learning called Boosted KNN (BK). The dataset has an imbalance issue in this research, with the stunting data at only 1% of the total dataset. Therefore, oversampling of the dataset is done by generating a random dataset based on the distribution of the data that are classified as the minority class. The results of elaborating on this oversampling are shown to be satisfying. Applying imbalanced data gives an average of 98% accuracy for all methods used; however, the F-1 score macro average is shown not optimal for each of the methods, with 51.95% for KNN, 52.45% for RF, and 53.55% for BK. After the data is balanced by oversampling, the F-1 score macro average for all methods substantially increases. The new results were 93.55% for KNN, 97.70% for RF, and 98.00% for BK, underscoring the critical role of addressing data imbalance in improving predictive accuracy.