<![CDATA[The increasing prevalence of imbalanced data in public health, particularly in nutritional surveillance among school-aged children, poses a significant challenge in accurately identifying cases of malnutrition, where the majority class (well-nourished children) often overshadows minority cases, leading to biased predictive outcomes. This issue is especially critical within the Free Nutritious Food (MBG) Program, where early detection of at-risk children is essential for effective intervention. Therefore, this study aims to develop a more reliable predictive model that can address class imbalance and improve the detection of malnutrition. To achieve this, the Synthetic Minority Oversampling Technique (SMOTE) was applied to balance the dataset by generating synthetic samples of the minority class, followed by classification using a Support Vector Machine (SVM). The model was evaluated using various train-test split ratios and assessed through multiple performance metrics, including accuracy, sensitivity, specificity, Cohen’s kappa, and AUC. The findings reveal that conventional SVM models trained on imbalanced data fail to detect malnutrition cases, resulting in zero sensitivity despite high accuracy. In contrast, the SMOTE-enhanced SVM significantly improves detection performance, achieving sensitivity above 90% and accuracy between 95% and 97%, with the optimal model obtained at an 70:30 split ratio. In conclusion, integrating SMOTE with SVM effectively overcomes class imbalance and provides a robust predictive framework for early identification of malnutrition in school-aged children.]]>
