Prediction Of Preterm Birth Using Ml On Electro Hysterography (Ego)Signals Form Physio Net

Abstract

Preterm labor is a critical issue in maternal healthcare that requires early and accurate prediction. Electrohysterogram (EHG) signals provide a non-invasive way to monitor uterine activity; however, these signals are noisy and datasets are highly imbalanced, which affects prediction performance. In this project, raw EHG signals are first processed using advanced denoising techniques, including Discrete Wavelet Transform (DWT), Savitzky–Golay filtering, and CEEMDAN, to improve signal quality. Relevant features are then extracted from the denoised signals. To handle class imbalance between term and preterm cases, SMOTE and BorderlineSMOTE are applied, and feature correlations are analyzed before and after balancing. Furthermore, different feature ranking techniques such as Mutual Information, ANOVA, and Recursive Feature Elimination (RFE) are applied to identify the most informative features. A comparative analysis is performed to evaluate the impact of denoising and feature selection methods on prediction performance. The proposed framework aims to improve the accuracy, robustness, and interpretability of preterm labor prediction, contributing to reliable clinical decision support systems.

Research Team