APPLICATION OF MACHINE LEARNING TECHNIQUE FOR THE PREDICTION OF NEONATAL MORTALITY USING MULTIPLE RISK FACTORS

Abstract

It is a common knowledge that one of the major cause of neonatal mortality is high-risk birth, which can be identified through risk factors. Though, there are many risk factors associated with neonatal mortality (i.eshort birth interval, pre-natal care etc), most interventions of government and other agencies target birth based on a single risk factor (i.e. poverty) even though most neonatal deaths are not from the targeted risk factor,thus, failing to curb the prevalence of early-life mortality. Hence, data from Nigerian Demographic and health Survey was gotten for this study and nine risk factors were used to predict neonatal mortality risk by applying Support Vector Machine algorithm to build a predictive model. KMeansSMOTE was used to solve the problem of class imbalance in the dataset, while model hyper-parameter tuning was appliedto the SVM model to get a better prediction. Neonatal mortality risk was estimated as a function of nine risk factors.  Risk factors chosen for the study were compared with four (4) risk factors from a previous study. The result gave a sensitivity of 78%, specificity of 44% and area under curve of 60% compared to using only four risk factors which has a sensitivity of 63%, specificity of 37% and area under curve of 50%. The result shows that having more risk factorsgives a considerable improvement by predicting more neonatal deaths.This will aid researchers and governments to identify more risk factorscausing neonatal deaths especially in Africa