Evaluation of Machine Learning Methods for Predicting the Risk of Child Mortality in South Africa

Dereje Danbe Debeko, Hawassa Univeristy
Reesha Kara, Rhodes University
Fidelia A. A. Dake, Regional Institute For Population Studies, University Of Ghana
Chodziwadziwa Kabudula, University of the Witwatersrand
Justin Dansou, Université de Parakou
Henry Wandera, PhD student
Chipo Mufudza, National University of Science and Technology

There have been extensive researches focused on child mortality in sub-Saharan Africa. But, the methods applied so far were based on the conventional regression analysis with limited prediction capability. Emerging methods in computational social science, particularly machine learning approach present opportunities to identify more features to facilitate accurate prediction of the risk of child mortality in sub-Saharan Africa. We evaluated different methods of machine learning techniques to develop the best model for predicting child mortality using training and test data from the National Income Dynamics Survey and District Health Barometer in South Africa. Logistic, Random Forest and XGBoost all show accuracy, sensitivity, and specificity of about 60%. Further analysis will be explored using data from different countries with different features.