Neonatal mortality remains unacceptably high in developing countries and the risk is greatest on the first day of life. A better perception of the causes responsible behinds the first-day neonatal mortality is a key to lessening this problem. This study assessed to predict and detect predicting factors of the FNM through different machine learning (ML) algorithms. The study data was based on FNM of 26145 children from the 2017-18 Bangladesh Demographic and Health Survey (BDHS). The Support Vector Machine (SVM) algorithm and chi-square test were used to extract predicting factors of the FNM. Prediction of IM was done using different ML models, for instance, decision tree (DT), random forest (RF), SVM, and logistic regression (LR). The performance of these techniques was evaluated via different parameters of confusion matrix, receiver operating characteristics (ROC) curve, and k-fold cross-validation. The study revealed that the prevalence of FNM was 3% (792 newborns out of 26145 children). Mother’s age at first birth, birth interval, region, religion, wealth index, child’s gender, birth order, and total children ever born were observed as significant predicting factors of the FNM in Bangladesh using the chi-square test. However, total children ever born, birth order number, father’s education, type of cooking fuel, exposure of media, wealth index, gender of the child, mother’s education, mother’s body mass index (BMI), and religion were the significant predicting factors of the FNM using the SVM method. To predict FNM in Bangladesh, though the LR model was performed better among all four ML algorithms based on the highest accuracy scores and the minimum standard error for the selected predicting factors using the SVM and chi-square test, the LR model failed to correctly predict the positive cases of FNM for sensitivity and precision. Needless to say, to predict the first-day neonatal mortality in Bangladesh for BDHS 2017-18 dataset, the SVM model was recommended (Accuracy = 0.9449, Sensitivity = 0. 0325, Specificity = 0.9745, Precision = 0.0396, area under the ROC curve (AUC) = 0.5227, k-fold accuracy = 0.9487) when the predicting factors will be identified using the SVM method, and the RF model (Accuracy = 0.9675, Sensitivity = 0. 0081, Specificity = 0.9986, Precision = 0.1538, area under the ROC curve (AUC) = 0.6461, k-fold accuracy = 0.9686) was recommended when the associated factors will be identified using the chi-square test. ML framework can be identified the significant predicting factors of the FNM, therefore may help the health-policymakers, stakeholders, and families to understand and prevent this severe public health problem.
 Save the Children. (2013). State of the world’s mothers: surviving the first day. London, UK. Available from: http://www.savethechildren.org/atf/cf/%7B9def2ebe-10ae-432c-9bd0-df91d2eba74a%7D/SOWM-FULL-REPORT_2013.PDF
 McGuire JW. Basic health care provision and under-5 mortality: a cross-national study of developing countries. World Dev. 2006; 34(3): 405–25
Financed by the National Centre for Research and Development under grant No. SP/I/1/77065/10 by the strategic scientific research and experimental development program:
SYNAT - “Interdisciplinary System for Interactive Scientific and Scientific-Technical Information”.