A Machine Learning Approach for Identifying and Predicting Risk Factors Related to Low Birth Weight in Newborn Children in Bangladesh

Background Low birth weight (LBW) is one of the principal causes of newborn death and several chronic diseases in developing countries of Southern Asia, particularly in Bangladesh. This paper primarily addressed the identification of risk factors for low birth weight in Bangladesh and the prediction of low birth weight in the country. Methods The secondary dataset of the Bangladesh Demographic and Health Survey (BDHS) 2022 which had 3400 respondents, was analyzed to identify the risk factors for low birth weight in Bangladesh. Feature importance was implemented using the XgBoost SHAP algorithm. Data transformation was applied using label encoding for all selected features. Nine machine learning (ML) classifiers (Random Forest (RF), Gradient Boosting (GB), Extra Trees (ET), Bagging, AdaBoost (AdaB), Logistic Regression (LG), Support Vector Classifier (SVC), Decision Tree (DT), and XGBoost) were trained and tested to classify low birth weight in Bangladesh. Accuracy, sensitivity, specificity, precision, F1-score, G-mean score, and area under the Receiver Operating Characteristic (ROC AUC) curve were calculated to evaluate the prediction model. Results With regards to all evaluation metrics of the prediction model, the XGBoost outperformed all other machine learning classifiers for LBW classification in Bangladesh, with the highest accuracy of 0.92, sensitivity 0.86, specificity 0.98, precision 0.98, F1-score 0.92, G-mean score 0.92, and ROC AUC curve 0.94. Conclusions The newborns with low birth weight were best classified by the XGBoost-based classifier, which also had the highest accuracy, precision, and G-mean score. The findings of this study suggest that to reduce and precisely predict low birth weight newborns in Bangladesh, an efficient, low-cost, and effective complementary method is needed.