Comparative Study of Machine Learning Algorithms to Predict Fetal Health Using Cardiotocography Data

Cardiotocography (CTG) monitors fetal heart rate and uterine contractions to assess fetal well-being during pregnancy. Monitoring fetal health is crucial throughout gestation. This paper presents a comparative study of 14 machine learning algorithms for predicting fetal health status, classified into three categories: normal, suspect or pathological. The evaluated models include Support Vector Machine (SVM), Attention Long Short-Term Memory (LSTM), Decision Tree (DT), XGBoost (XGB), Cost-Sensitive Random Forest (CSRF), Self-Supervised PCA + Logistic Regression, Graph Neural Networks (GNNs), K-Nearest Neighbour (KNN), Gradient Boosting (GBM), Voting Classifier, Random Forest (RF), Multi-Layer Perceptron (MLP) and One-Dimensional Convolutional Neural Network (1D-CNN). To address the inherent class imbalance and curse of dimensionality, we employed the Random Oversampling technique and SelectFromModel using random forest respectively, resulting in a more representative dataset. Key performance metrics including precision, accuracy, recall and F'l-Score were measured and tabulated to provide a clear comparison of each model's performance. Of all the algorithms evaluated Random Forest (RF), Cost-Sensitive Random Forest (CSRF) and XG Boost (X G B) were the best performers, all achieving similar training and testing accuracies of 99.9% and 99.8% respectively in predicting fetal health status.