Enhancing Stunting Prediction for Indonesian Children Using Machine Learning with SMOTE Data Balancing

Stunting is a significant health issue in Indonesia, affecting the growth of children under five due to chronic malnutrition. Traditional methods for early identification of at-risk children often fall short, highlighting the need for advanced predictive models like machine learning (ML). This study compares the performance of support vector machine (SVM) and Decision Tree algorithms using data from Bandarharjo Community Health Centers. Initial results show poor performance for SVM models with linear and polynomial kernels, achieving F1-scores between 0% and 54%. The decision tree algorithm performed slightly better with an F1-score of 64%. To improve detection accuracy, the Synthetic Minority Over-sampling Technique (SMOTE) was applied as a data balancing technique to address class imbalance. After applying SMOTE, the decision tree achieved an F1-score of 97%, proving to be the most effective model. The SVM with the radial basis function (RBF) kernel also improved significantly, achieving an F1-score of 94%. These findings demonstrate that data balancing techniques like SMOTE are crucial for enhancing the accuracy and effectiveness of ML models in detecting stunting, enabling more timely and accurate health interventions.