Early Detection of Childhood Malnutrition Status in Indonesia: An Imbalance-Handling and Explainable AI Approach

Abstract

The Double Burden of Malnutrition is a serious health problem for communities worldwide, especially for children under 5 years of age, who are vulnerable to nutritional problems. If children suffer from nutritional problems, this can affect their cognitive and physical health, as they are still in the growth phase. This study was conducted to study and test a machine learning-based framework to predict the nutritional status of toddlers using datasets from the Sawah Besar Community Health Center. Datasets include demographic and anthropometrics indicators, based on age, gender, height, weight, and weight for height index. To spot class imbalance found in healthcare datasets, the Synthetic Minority Oversampling Technique (SMOTE) was applied to generate a balanced training dataset. Four supervised machine learning algorithms AdaBoost, LightGBM, CatBoost, and Logistic Regression were evaluated using both balanced and imbalanced datasets. Among other models, LightGBM with balanced data had achieved the best performance report with an accuracy of 89.5%, a precision of 90.1%, a recall of 89.5%, and an F1-score of 89.8%. In addition, Shapley Additive exPlanations (SHAP) were utilized to identify the most influential features to model predictions and to enhance model interpretability. Based on the Shapley Additive Explanations (SHAP), the results indicate that anthropometric variables are the most contributing features to the prediction model. Furthermore, this study supports Sustainable Development Goal 3, Good health and well-being.