Application of machine learning methods for predicting under-5 mortality determinants in Bangladesh: evidence from the 2017-18 Bangladesh Demographic and Health Survey

Abstract

Abstract Background: Under-5 mortality is a vital social indicator of a country's development and long-term economic viability. The most underlying factors contributing under-5 mortality is a concern in developing countries like Bangladesh. There has been extensive research conducted on under-5 mortality. The prevailing approach employed thus far primarily relies on traditional logistic regression analysis, which have demonstrated limited predictive effectiveness. Advance Machine Learning (AML) methods provide accurate prediction of under-5 mortalities. This study utilized Machine Learning techniques to forecast the mortality rate among children under the age of five in Bangladesh. Methods: The data for the study were drawn from the Bangladesh Demographic Health Survey 2017–18 data. Python version 3.0 software was utilized to implement and evaluate various Machine Learning (ML) techniques, including Decision Tree (DT), Random Forest (RF), Logistic Regression (LR), K-Nearest Neighbors (KNN) and Support Vector Machine (SVM). Boruta algorithm for selecting best features by using Boruta packages of R programming language. Furthermore, the SPSS Version 17 was used for analyzing conventional methods. Various matrices, like confusion matrix, accuracy, precision, recall, F1 score and the Area Under the Receiver Operating Characteristic Curve (AUROC) was utilized as a metric to assess the effectiveness or performance of predictive models. Results: We opted for t2xhe Random Forest (RF) model is the best predictive model of under-5 mortality in Bangladesh with accuracy (95.97%), recall (11%), precision (40%), F1 score (18%), and AUROC (75%). Our predictive models showed that Currently breastfeeding, Wealth index, Religion, Birth order number, Number of household members, Place of delivery, Type of toilet facility, Type of cooking fuel are the 8 top determinants of under-5 mortality in Bangladesh. Conclusions: Machine Learning methods were utilized to create the most optimal predictive model enabling the classification of hidden information that remained undetectable through traditional statistical methods. In our Study the Random Forest model was the best models for predicting under-5 mortality in Bangladesh.