This study investigates the application of SMOTE (Synthetic Minority Over-sampling Technique) to address class imbalance in children’s nutritional status datasets, focusing on two indicators: BB/U (Weight-for-Age) and BB/TB (Weight-for- Height). The goal is to enhance the predictive performance of machine learning models, particularly in classifying underrepresented nutritional categories. K-Nearest Neighbors (KNN) and Random Forest were employed to evaluate SMOTE’s effectiveness. The results reveal significant improvements in recall for minority classes. For KNN, testing accuracies reached 96.66% for BB/U and 93.58% for BB/TB, with enhanced recall values for minority categories. Random Forest demonstrated superior performance with cross-validation accuracies of 97.59% for BB/U and 94.79% for BB/TB, achieving balanced classification across major and minor classes. The dual use of BB/U and BB/TB as target columns proved crucial for a comprehensive assessment of nutritional status, as each captures different dimensions of child growth. Additionally, key features such as gender and prior weight status were found to significantly influence model predictions. By improving the ability to detect at-risk groups, this study offers actionable insights to support more precise and data-driven nutritional interventions. The findings provide valuable guidance for policymakers and healthcare professionals in Indonesia, contributing to more effective strategies to combat childhood malnutrition and promote equitable health outcomes. These results highlight the potential of machine learning techniques, when combined with SMOTE, to address public health challenges in a robust and scalable manner.