Machine Learning Algorithm for Analysing Infant Mortality in Bangladesh

The study aims to investigate the potential predictors associated with infant mortality in Bangladesh through machine learning (ML) algorithm. Data on infant mortality of 26145 children were extracted from the latest Bangladesh Demographic and Health Survey 2017-18. The Boruta algorithm was used to extract important features of infant mortality. We adapted decision tree, random forest, support vector machine and logistic regression approaches to explore predictors of infant mortality. Performances of these techniques were evaluated via parameters of confusion matrix and receiver operating characteristics curve. The proportion of infant mortality was 9.7% (2523 out of 26145). Age at first marriage, age at first birth, birth interval, place of residence, administrative division, religion, education of parents, body mass index, gender of child, children ever born, exposure of media, wealth index, birth order, occupation of mother, toilet facility and cooking fuel were selected as significant features of predicting infant mortality. Overall, the random forest (accuracy = 0.893, precision = 0.715, sensitivity = 0.339, specificity = 0.979, F1-score = 0.460, area under the curve: AUC = 0.6613) perfectly and authentically predicted the infant mortality compared with other ML techniques, including individual and interaction effects of predictors. The significant predictors may help the policy-makers, stakeholders and mothers to take initiatives against infant mortality by improving awareness, community-based educational programs and public health interventions.