Unveiling socio-demographic determinants of low birth weight using machine learning techniques.

Low birth weight (LBW) poses significant challenges to child survival, contributing to increased rates of mortality and morbidity, and has long-term adverse effects on overall health. The persistently high prevalence of LBW in low- and middle-income countries, including Bangladesh, reflects underlying health disparities. Despite recent improvements, Bangladesh still reports a notable LBW rate of 14.5%, indicating persistent maternal and child health concerns. Various socio-demographic factors influence birth weight, necessitating a comprehensive investigation into their contributions. This study aims to identify the key determinants of LBW and develop a machine learning-based predictive model to assess vulnerable mothers of having LBW babies based on risk factors associated with birth weight. Data for this study were obtained from the Bangladesh Demographic and Health Survey (BDHS) 2022, which encompassed 2,621 women (excluding missing cases) and 8,784 women (including missing cases). Several machine learning algorithms, including logistic regression, Naïve Bayes, k-nearest neighbors (KNN), random forest, support vector machine (SVM), Lasso regression, regression tree, neural networks, XGBoost, AdaBoost, and decision tree classifiers, were employed to analyze the risk factors. Model performance was evaluated using a train-test split approach and 10-fold cross-validation, with accuracy, precision, recall, F1-score, R² score (only for the regression model), and mean squared error (MSE) as assessment metrics. The findings indicate that 'Age at first birth' and 'Education Level' emerged as the most influential predictors of LBW, while AdaBoost demonstrated the highest predictive accuracy among the applied models. The findings of this study might make significant contributions in identifying vulnerable mothers giving birth to children with LBW and making policies highlighting risk factors responsible for LBW to reduce the frequency of LBW.