Identifying risk factors for newborn outcomes using artificial neural networks

Abstract

The goal of this thesis is to identify the risk factors for caesarean delivery, neonatal mortality and low Apgar score using Artificial Neural Networks (ANNs). The medical domain of interest used is the perinatal database provided by the Perinatal Partnership Program of Eastern and Southeastern Ontario (PPPESO). The ability of the ANNs to generate strong predictive model with the most influential variables was tested. Different ANN techniques for weight extraction and determining the importance of each input variables were applied. The thesis used feedforward ANNs trained by the backpropagation algorithm, as this is a widely used ANN in medical applications. Finally, minimal sets of variables (risk factors) that are important in predicting each outcome without degrading the ANN performance were identified.