Algorithmic Midwifery: Benchmarking Machine Learning Models for Intelligent Fetal State Classification

The fetal condition is of great significance in the prenatal care and minimizing the negative birth complications. This paper presents a baseline model, which is known as Algorithmic Midwifery, that implements classical machine learning algorithms to classify the intelligent fetus state through the Cardiotocography (CTG) signals. To classify fetal conditions into the normal, suspect, and pathological classes, three popular models, including K-Nearest Neighbours (KNN), Support Vector Machine (SVM), and Random Forest (RF), were used. Very few preprocessing steps such as median-based imputation of missing values, scaling of the features, and label encoding were used so that the original data integrity was not lost. Grids of hyperparameters were optimized with grid search and 5-fold cross-validation, and the performance of the model was evaluated with the accuracy, precision, recall, F1-score and the analysis of the confusion matrix. SVM had the best performance and reported an overall accuracy of 92.4 which exhibited balanced accuracy-recall behavior. RF offered an important amount of feature importance scores, which are easier to interpret, whereas KNN was used as a simple and efficient baseline reference. Misclassifications that are present between Suspect and Pathological classifications underline the necessity to refine them further. All in all, this framework provides a solid classical-machine-learning baseline of fetal state classification and is a solid reference point of future deep learning methods, with practical implications of creating clinically-reliable decision-support systems in practice.