Machine Learning And Filtering Techniques For Artefact Removal In EDA Signals During Fetal Movement Monitoring

Abstract

Electrodermal Activity (EDA) signals are valuable for monitoring stress and emotional states, but are often corrupted by motion and environmental artefacts. This study compares traditional filtering techniques (Butterworth, wavelet) with machine learning (Isolation Forest, Support Vector Machines (SVM), XGBoost) for artefact removal. Using both real and synthetic noise-contaminated datasets, the models were evaluated using Signal-to-Noise Ratio (SNR), Mean Squared Error (MSE), and correlation with clean reference signals. The results indicate that while conventional filtering techniques are effective for reducing simple noise components, deep learning approaches, particularly DCA, demonstrate superior denoising performance. These findings highlight the potential of AI-based methods for reliable, real-time EDA signal processing in wearable healthcare systems, especially for applications such as fetal movement monitoring.