A Fusion of Wavelet-based and Unsupervised Machine Learning Method for Artifacts Removal in Electrodermal Activity Signal

Abstract

The fast tempo of modern society has brought people a series of emotional changes and mental pressures. Therefore, research have emerged to help people pay attention to and regulate their mental health. Physiological signals are used in studies from different fields to monitor and detect emotional change and stress. Electrodermal activity (EDA) is such a physiological signal that can reflect changes in skin conductivity when people's emotions change. The nature of neuromodulation makes such changes not easily controlled by people subjectively so EDA is an ideal emotion and stress monitoring indicator. Especially with the popularity of wearable devices in the market, wearable devices with built-in EDA sensors will be more competitive for the functions that help people regulate their mental health. However, since the EDA sensor usually acquires signals through fingers, palms, or wrists, artifacts will inevitably be generated when users move their hands or arms, and the artifacts will affect the accuracy of emotional change detection. As a result, removing artifacts in EDA signals is a challenging and important topic. In this work, multiple signal processing methods are applied to realize the objective of removing artifacts in the EDA signals from the AMIGOS dataset. The results show that the proposed method is promising and has the potential to be utilized in real-time EDA signal processing and emotional change detection applications.