Implementation of a Predictive Model for Skilled Child Delivery Service use in Afghanistan

Abstract

Institutional delivery during childbirth is essential to reduce both maternal and child mortality. Nevertheless, in Afghanistan, which has a high rank of maternal and child mortality around the globe, the number of childbirth attended by skilled birth attendants (SBAs) in health facilities remains extremely low. Therefore, the ability to predict the skilled child delivery service use is helpful and an excellent preventive measure. This study aims to develop a web-based skilled child delivery service use predictive model using data mining classification algorithms and identify the most suitable classifier among the four well-known machine learning algorithms. These are Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and PART rule induction. Waikato Environment for Knowledge Analysis (WEKA) version 3.8.4 was used to develop optimal models. The dataset used is the Afghanistan Demographic and Health Survey (AfDHS). The classification in this study comprises two categories, ‘Skilled delivery’ and ‘No skilled delivery’. Preparation of the dataset is carefully done to ensure well-balanced samples in each category. The validation of the predictive models is assessed by means of Accuracy, Precision, Recall, and area under Receiver Operating Characteristics (ROC) curve. The study found that Random Forest is the best classifier with an accuracy, precision, recall, and the area under ROC of 84.23%, 84.40%, 84.20% and 91.70% respectively. Subsequent to developing an optimal predictive model, we relied on this model to develop a web-based mobile application system for skilled child delivery service use prediction. Thus, the result can help decide targeted interventions for pregnant women to ensure skilled assistance at child delivery.