Fetal health screening model and element analysis

Abstract

The United Nations Sustainable Development Goals has mentioned to reduce child mortality. That is also a crucial indicator of human progress. The UN hopes that all countries will eradicate preventable deaths of newborns at the end of 2030. Cardiotocogram (CTG) can be used to identify in-danger women during pregnancy. The aim of this article is to apply machine learning algorithm techniques on CTG data to ensure fetal well-being. CTG data of 2126 samples and 22 variables were obtained from the CTG exams on Kaggle. Two different classification models were trained through the data. In order to predict Normal, Suspect, and Pathological fetal states, each class had its own sensitivity, precision and F1 score. Each model has its overall accuracy. Determined by obstetricians interpretation of CTG, Normal state accounted for 57%, Suspect state accounted for 23% and Pathological state accounted for 20%. The classification models generated by Logistic Regression and Random Forest to predict the suspect and pathological state of the fetus by tracing CTG. They had high precision of 86% and 94% respectively. However, the classification model developed by Random Forest had higher prediction accuracy for a negative fetal outcome. Healthcare workers without professional training in low-income countries have the opportunity to utilize this model for the purpose of prioritizing pregnant women in hard-to-reach regions, ensuring they receive timely referrals and appropriate follow-up care.