Autonomous Fetal Distress Detection based on performance of Machine learning algorithms

Abstract

Cardiotocography (CTG) is a medical tracking procedure used to assess the well-being of the foetus by monitoring the patterns of its heart ratein response to the mother's signal. Although CTG is predominantly used tool to monitor and detect the health of the foetus, the increase in the results of false alarm rates due to visual deciphering highly constitute to unnecessary operative delivery or delayed intrusion. A novel automatic process is proposed here for early diagnose and detect of foetus abnormality using machine learning approach. The dataset is taken from Cardiotocography UCI repository which holds 2126 instances with normal, suspect, and pathological (N, S, P) classes obtained from measurements of uterine contraction (UC) and fetal heart rate (FHR)features. Following feature scaling and normalization, the feature data is fed into machine learning models like Naive Bayes, Support Vector Machine, k-Nearest Neighbor, and Random Forest techniques to classify the imbalanced data into multiclass categories N, S, P.The various performance metrics were calculated for four algorithms and the results show that Random Forest within computational time of 6.32shas obtained overall Accuracy of 90.82%, weighted F₁ score of 91.24%, mean MCC (Mathews Correlation Coefficient) of about 74%Mean Kappa Score of 72.82% and Averaged Area under the ROC of 0.8766 which is better when compared to other algorithms. Hence Random Forest method can be used to autonomously detect the fetaldistress during pregnancy.