<![CDATA[Newborns are classified as having low birth weight (LBW) if they weigh less than 2,500 grams (5.5 pounds), which poses risks for stunted growth, developmental delays, and long-term health issues such as cardiovascular disease. This study develops an XGBoost based machine learning model for low birth weight classification using the 2017 Indonesia Demographic and Health Survey (IDHS). To address severe class imbalance, various resampling techniques including undersampling, oversampling, hybrid methods, and algorithm level class reweighting are applied, along with a modified random oversampling method (M-ROS). SHAP and XGBoost’s built in feature importance are then used to identify key predictors of low birth weight in the Indonesian context. A total of 16,340 records with 42 preprocessed features were used in this study. The dataset was split into 70 percent for training and 30 percent for testing, and stratified k-fold cross-validation with random search was employed to identify the optimal hyperparameters. Among the resampling strategies evaluated, random oversampling, M ROS, ENN with class reweighting, and class reweighting without any resampling were able to improve recall. Of these approaches, class reweighting without resampling achieved the highest recall at 36.63 percent, while ENN combined with class reweighting produced the highest accuracy at 84.72 percent. Key predictors influencing the classification of low birth weight included the number of antenatal care visits, multiple pregnancy, and place of delivery. The findings demonstrate that machine learning can be effectively utilized to predict low birth weight and to identify influential contributing factors.]]>
