EHG FOR PREDICTION OF LABOUR IN PREGNANT WOMEN USING OPTIMIZED MACHINE LEARNING MODELS

Abstract

Early prediction of preterm labour is vital to reduce neonatal morbidity and mortality. This study explores the potential of electrohysterogram (EHG) signals for classifying term and preterm uterine activity using optimized machine learning models. The Term–Preterm EHG dataset from PhysioNet was employed. After band-pass filtering and segmentation, a set of time–frequency and nonlinear features was extracted from each EHG window. An improved Firefly Algorithm (IFA) was used for feature optimization before classification with Support Vector Machine (SVM), Gradient Boosting Classifier (GBC), k-Nearest Neighbour (KNN), Naïve Bayes, and Decision Tree models. To avoid data leakage, patient-level 5-fold cross-validation was adopted. Performance was evaluated using accuracy, sensitivity, specificity, precision, ROC–AUC, and confusion matrices. The optimized IFA–SVM achieved the highest mean accuracy of 97.6 ± 0.8% and ROC–AUC of 0.983, outperforming other classifiers (p < 0.05). Feature importance analysis indicated that entropy and spectral energy were the most discriminative features between term and preterm contractions. These findings suggest that EHG-based modelling, when rigorously validated, may assist clinicians in identifying pregnancies at risk of preterm labour. Further clinical trials are required for real-world validation.