Deep Learning-based Multimodal Analysis of EEG-EDA for Human Responses to Construction Noise Exposure

초록

This study aims to evaluate the effectiveness of multimodal fusion of human physiological responses subjected to various levels of construction noise exposure. Construction noise is a common external stressor known to affect human physiology and perception, affecting cognitive performance. Electroencephalography (EEG) and electrodermal activity (EDA) offer complementary insights into how the brain and autonomic nervous system respond to noise. EEG reflects brain activity related to perception and attention, while EDA measures sympathetic nervous systems associated with arousal and stress responses. To assess these complex physiological reactions, this study employs a multimodal deep learning framework that integrates responses to construction-related noise of different intensities. Two types of artificial construction noise were used: impulsive noise from pile drivers and tonal noise from earthquake augers. Twenty-five participants were exposed to three simulated noise levels (40 dB, 60 dB, and 80 dB), during which both EEG and EDA signals were recorded simultaneously. Separate deep learning models: Long short-term memory for EDA and convolutional neural networks for EEG were employed to analyze preprocessed signals independently. In the decision-level fusion phase, the output predictions from each modality-specific deep learning model were combined by utilizing a fusion strategy, weighted averaging, to arrive at a final prediction of the construction noise exposure level. Initial results indicate the effectiveness of the decision-level multimodal deep learning method, with the integrated model exhibiting superior capability to differentiate among various intensities of construction noise compared to unimodal models. © 2025, Korea Institute of Construction Engineering and Management. All rights reserved.

키워드