Since people in contemporary society spend most of their time interacting with the built environment, there is a growing need to thoroughly understand the quality of human-built environment interaction to improve quality of life. Recent wearable electrodermal activity (EDA) sensing has shown the potential to meet this need by continuously, less invasively, and less laboriously monitoring individuals' stress levels as an important dimension of the quality of interaction with the built environment. However, analyzing EDA to detect stress is still challenging due to significant intrinsic and extrinsic noises in EDA collected by a wearable biosensor in the field. Although several denoising methods have been proposed based on differences in signal characteristics between noises and desired EDA signals evoked by sources of interest (e.g., stress), these methods do not address intrinsic respiration noise due to similarities in the signal characteristics of respiration noise and desired EDA signals. To address this issue, the authors propose a denoising method that references simultaneously collected photoplethysmography (PPG) as a respiration noise-correlated signal to attenuate respiration noise as well as extrinsic noises. The performance of the proposed method was compared with advanced benchmark denoising methods using 25 subjects' stress data collected in the field. As a result, stress metrics calculated from EDA denoised using the proposed method were statistically more valid and reliable than ones from EDA denoised by benchmark denoising methods. Accordingly, machine learning models trained by having the stress metrics as features showed statistically higher accuracy with EDA denoised by the proposed method than by benchmark denoising methods. These results show that the proposed method can improve stress measurement using EDA by attenuating both intrinsic respiration noise and extrinsic noise. The finding contributes to the body of knowledge by demonstrating that intrinsic noise with signal characteristics indistinguishable from desirable signals can be suppressed by referencing another noise-correlated signal effortlessly acquired using multimodal wearable biosensors. This new knowledge will facilitate the application of wearable EDA sensing devices to continuously, less invasively, and less laboriously measure people's stress in their daily interactions with the built environment. Using wearable-based stress measurement, urban managers can detect and address environmental stressors in the built environment in a more scalable manner, thereby more effectively improving the quality of interaction between humans and the built environment.
|Original language||English (US)|
|Journal||Journal of Computing in Civil Engineering|
|State||Published - Nov 1 2020|
All Science Journal Classification (ASJC) codes
- Civil and Structural Engineering
- Computer Science Applications