Evaluation of support vector machines and random forest classifiers in a real-time fetal monitoring system based on cardiotocography data

In this paper, we compare methods for evaluating the fetal state prediction based on Cardiotocography (CTG) data. Antepartum Fetal Monitoring provides information that can be used to predict the state of the fetus during labor to indicate the risk of a fetal acidosis (low blood pH from low oxygen levels). The effectiveness of these predictions is evaluated in a real-time clinical decision support system and extracts other features that can provide further information regarding the fetal state. This research differs from previous work in that all three fetal states (normal, suspect and pathological) are considered. The paper discusses the importance of machine learning in providing assistance for the obstetricians in `suspect' cases. Results show that both Support Vector Machines and Random Forests had over 96% accuracy when predicting fetal outcomes, with SVM performing slightly better for suspect cases.