Machine learning analysis of the association between fertility and under five mortality in Bangladesh using BDHS 2022 data

Under-five mortality is a key indicator of child health, reflecting healthcare quality, nutrition, and socio-economic conditions. In Bangladesh, under-five mortality remains a public health concern despite substantial declines, with fertility-related factors potentially contributing to child survival. Traditional logistic regression may not adequately capture complex and nonlinear relationships, whereas machine learning methods can capture such patterns and provide complementary predictive insights. This study examined the association between fertility-related factors and under-five mortality using traditional logistic regression and machine learning approaches with nationally representative BDHS 2022 data. Data were obtained from the Bangladesh Demographic and Health Survey (BDHS) 2022. Ten machine learning models Logistic Regression, Support Vector Machine, K-Nearest Neighbors, Random Forest, Decision Tree, Gradient Boosting, AdaBoost, Neural Network, XGBoost, and LightGBM were implemented in Python, while complex-sample logistic regression was conducted using SPSS Version 25. Model interpretability was assessed using SHAP values and partial dependence plots (PDPs). Model performance was evaluated using accuracy, precision, recall, F1-score, Matthews Correlation Coefficient (MCC), Cohen’s Kappa, and the area under the receiver operating characteristic curve (AUROC). Decision Tree demonstrated the best overall performance for identifying under-5 mortality, achieving a mortality precision of 15.25%, recall of 52.94%, F1-score of 0.2368, MCC of 0.2424, Cohen’s Kappa of 0.1973, and AUROC of 0.8045. Gradient Boosting achieved the highest mortality recall (55.88%). Key predictors included household size, total children ever born, currently breastfeeding status, and parental education. SHAP and PDP analyses identified important contributions from fertility-related, household, and socio-economic factors to model predictions. Higher parity, measured by the total number of children ever born, was significantly associated with increased odds of under-five mortality. Current breastfeeding status was inversely associated with mortality; however, this finding should be interpreted cautiously because of potential temporal ambiguity. Machine learning models complemented complex-sample logistic regression by identifying important fertility-related, household, and socioeconomic predictors and capturing potentially nonlinear patterns in the data. These findings provide evidence to inform further research and public health planning aimed at reducing under-five mortality in Bangladesh.