Lakhasly

Significant intrinsic and extrinsic noises in EDA collected by a wearable biosensor in the field hinder the application of EDA- based stress measurement for understanding the interaction be- tween humans and the built environment.Also, the validity and reliability of the stress metrics (i.e., BCC and ICC) calculated from EDA denoised by the pro- posed method were statistically higher than those from EDA denoised by advanced benchmark methods.The proposed denoising method contributes to monitoring and improving the quality of the human-built envi- ronment interaction by enabling wearable EDA sensors to collect high-quality signals and accurately measure stress from people's daily outings.Consequently, classi- fication models based on stress metrics showed statistically higher accuracy with EDA denoised by the proposed method than by the benchmark methods.This finding demonstrates that intrinsic noise with signal characteristics similar to those of desired signals can be alleviated by referencing other signals readily acquired using multimodal wearable biosensors.Then intrinsic respiration noises are detected and attenuated using a subject- independent machine learning model that detects noise-inducing irregular respiration from PPG collected at the same time as EDA.The proposed method first attenuates extrinsic noises by applying sev- eral filters (e.g., high-pass filter and wavelet filter).These results indicate that the proposed denoising method can improve stress measurement using EDA by attenuating both intrinsic respiration noise and extrinsic noise in EDA.The subject-independent irregular respiration classifier trained by applying a MDAN showed 0.849 accuracy in LOSOCV.

Significant intrinsic and extrinsic noises in EDA collected by a
wearable biosensor in the field hinder the application of EDA-
based stress measurement for understanding the interaction be-
tween humans and the built environment. To address this issue, this
study proposed a denoising method that references PPG to alleviate
intrinsic respiration noise as well as extrinsic noise in EDA. The
proposed method first attenuates extrinsic noises by applying sev-
eral filters (e.g., high-pass filter and wavelet filter). Then intrinsic
respiration noises are detected and attenuated using a subject-
independent machine learning model that detects noise-inducing
irregular respiration from PPG collected at the same time as
EDA. To test the denoising performance of the proposed method,
the authors collected data in the lab with 10 subjects and in the field
with 25 subjects. The subject-independent irregular respiration
classifier trained by applying a MDAN showed 0.849 accuracy
in LOSOCV. Also, the validity and reliability of the stress metrics
(i.e., BCC and ICC) calculated from EDA denoised by the pro-
posed method were statistically higher than those from EDA
denoised by advanced benchmark methods. Consequently, classi-
fication models based on stress metrics showed statistically higher
accuracy with EDA denoised by the proposed method than by the
benchmark methods. These results indicate that the proposed
denoising method can improve stress measurement using EDA by attenuating both intrinsic respiration noise and extrinsic noise in
EDA. This finding demonstrates that intrinsic noise with signal
characteristics similar to those of desired signals can be alleviated
by referencing other signals readily acquired using multimodal
wearable biosensors. The proposed denoising method contributes
to monitoring and improving the quality of the human–built envi-
ronment interaction by enabling wearable EDA sensors to collect
high-quality signals and accurately measure stress from people’s
daily outings.