Statistical Insights Into Machine Learning Models for Predicting Under-Five Mortality

It examines the use of machine learning models on mortality prediction focusing on feature selection, data balancing, and ensemble methods. In the present study, the Under-Five Mortality Dataset contains records for sex of the child, type of birth, place of residence, region of residence, educational status of the mother, household food security, birth intervals, antenatal care, place of delivery, and maternal age. Some Machine Learning techniques, which include Random Forest and Stacking Classifier, together one of the strongest ensemble methods, are used in modeling under-five mortalityK-Best is an approach for feature selection which would ensure that the minimal sufficiently informative set of features is retained, improving the model performance. It achieves class balancing through the use of synthetic minority oversampling technique (SMOTE) presentation for mortality- and non-mortality cases. The current analysis is going to lead to determining important statistical results concerning important under-5 deaths and predictive ability of machine learning models in public health application. It is shown that better analyses can be done on which populations have the greatest risk and thus inform health policies that may cure various different causes of under-five mortality in the different regions.  

  Keywords: These are graphical representations for the meanings associated with the terminologies such as under-five mortality, machine learning, random forest, stacking classifier, K Best, SMOTE, synthetic oversampling, and public health, and those pertaining to predictive modeling, child health, maternal health, class imbalance, data balancing, and mortality prediction.