Application of machine learning methods in analysis of infant mortality in Rwanda: Analysis of Rwanda Demographic Health Survey 2014-15 dataset

Abstract

Extensive research on infant mortality (IM) exists in developing countries; however, most of the methods applied thus far relied on conventional regression analyses with limited prediction capability. Machine learning (ML) methods is new method used to provide accurate prediction of the factors associated with IM; however, there is no study conducted using ML methods in Rwanda. This study used ML methods to determine factors associated with IM and building its predictive models. A cross-sectional study design was conducted using 2014-15 Rwanda Demographic and Health Survey. Python software was employed to apply ML methods through Logistic Regression, Random Forest, Decision Tree and Support Vector Machine. Multivariate logistic regression was employed as a traditional method. Evaluation metrics methods specifically confusion matrix, accuracy, precision, recall, F1 score, and Area under the Receiver Operating Characteristics (AUROC) were used to evaluate the performance of predictive models. Marital status, maternal education, wealth index, sex of child and birth interval was statistically significant factors associated with infant mortality. By applying ML methods, our results revealed that random forest model was best predictive model of infant mortality with model accuracy (84.29%), recall (91.33%), precision (80.31%), F1 score (85.46%) and AUROC (84.20%); followed by decision tree model with model accuracy (83.02%), recall (90.97%), precision (78.96%), F1 score (84.67%) and AUROC(82.94%), followed by super vector machine with model accuracy (68.62%), recall (74.94%), precision(66.97%), F1 score (70.73%) and AUROC (68.55%) and last was logistic regression with low accuracy of prediction (61.49%), recall (61.05%), precision (62.15%), F1 score (61.59%) and AUROC (61.50%) compared to other predictive models. In developing a predictive model, ML methods are used to classify certain hidden information that could not be detected by traditional statistical methods. Random forest was classified as the best classifier to be used for the predictive models of infant mortality.