Machine Learning Techniques to Identify Determinants of Infant and Child Mortality Based on Ethiopia Demographic and Health Surveys (EDHS)

Abstract

The Ethiopian government doing for the past two decades for attaining millennium development goals agenda for preventing childhood mortality by improving the child health’s to change the country image to the rest of the world in reduction of childhood mortality. This study contributes some values in the improvement of childhood health by analyzing the determinants infant and child mortality by using machine learning techniques. Different reports indicate that the distribution of childhood mortality differs in the world. According to the United Nations Inter-Agency Group, 2017 report the global under-five mortality rate declined by 56 percent starting from 1990 to 2016. In 1990 the deaths per 1,000 live births were 93 percent and in 2016 counted 41 percent. Ethiopian demographic and health survey 2016 report of 2017 indicates that for the 5 years preceding the survey, the under-5 mortality rate was 67 deaths per 1,000 live births, and the infant mortality rate was 48 deaths per 1,000 live births. This study aimed to identify factors and developed predictive models using four supervised machine learning techniques namely C5.0 Decision tree, Random Forest, Support Vector Machine and Naïve Bayes algorithms using the 2016 EDHS dataset of 10,641 records. K-fold cross-validation, performance measures accuracy and Kappa and confusion matrix metrics used for performance measurements of models. SMOTE sampling method was used as the best sampling techniques. The child mortality rate was 60 deaths per 1000 live births. Number of under-five child in household (AOR = 16.08, (95% CI [13.3,19.44]), type of place of residence (AOR = 0.73, 95% CI [0.5,1.08]), source of drinking water (AOR = 0.97, [0.94,0.99]), wealth index difference (AOR = 0.97, [0.89,1.06]), family plan use (AOR = 1.03, 95% CI [0.99,1.07]), breastfeeding (AOR = 2.58, 95% CI [2.12,3.14]), place of delivery (AOR = 1.0062, 95% CI [0.9939,1.0187]), birth order number (AOR = 1.73, 95% CI [1.35,2.22]), and, preceding birth interval (AOR = 1.0075, [1.0017,1.0133]) were found to be determinants for child mortality. In this research work, Random Forest and C5.0 decision tree model had performance , accuracy (95.93%), (96.15%), sensitivity (65.26%), (65.78%), specificity (97.56%), (98.06%), Positive Predictive Value (PPV) (65.96%), (68.31%), Negative Predictive Value (NPV) 97.80%), (97.84%) respectively. The result obtained could support child health intervention programs in Ethiopia and used as a reference for the researcher.