Dolphin
Senior Member (Voting Rights)
https://link.springer.com/chapter/10.1007/978-3-032-29456-2_18
Part of the book series: Lecture Notes in Computer Science ((LNAI,volume 16707))
Included in the following conference series:
In an experiment at an actual hospital, physiological data: electrocardiogram (ECG), electrodermal activity (EDA), blood volume pulse (BVP) and skin temperature were acquired. A machine learning model was constructed using an index that integrated subjective fatigue (VAS) and objective functional decline (critical fusion frequency: CFF) via Principal Component Analysis (PCA) as the ground truth label. In constructing the model, physiological indices effective for estimating fatigue states were extracted by calculating feature importance using LightGBM. The analysis revealed that physical fatigue strongly depends on recent heart rate indices, whereas mental fatigue strongly depends on long-term EDA indices, confirming differences in physiological characteristics according to the kind of fatigue.
The constructed classification model showed high accuracy (Accuracy: 0.634) against measured values, demonstrating that this method is effective as a fundamental technology for a practical Fatigue Risk Management System (FRMS).
This is a preview of subscription content, log in via an
- Home
- Engineering Psychology and Cognitive Ergonomics
- Conference paper
Fatigue Estimation Using a Wearable Device Aimed at Application to Fatigue Risk Management
- Conference paper
- First Online: 22 June 2026
- pp 256–268
Part of the book series: Lecture Notes in Computer Science ((LNAI,volume 16707))
Included in the following conference series:
- 43 Accesses
Fatigue Risk Management (FRM), one of the factors contributing to human error in medical settings, is an urgent issue. However, continuous monitoring in actual work environments has been difficult because conventional evaluations based on self-reports or intermittent tests interfere with work duties. This study proposes a method to estimate the fatigue state of nurses during work continuously and objectively using multimodal physiological measurement with a wearable device.In an experiment at an actual hospital, physiological data: electrocardiogram (ECG), electrodermal activity (EDA), blood volume pulse (BVP) and skin temperature were acquired. A machine learning model was constructed using an index that integrated subjective fatigue (VAS) and objective functional decline (critical fusion frequency: CFF) via Principal Component Analysis (PCA) as the ground truth label. In constructing the model, physiological indices effective for estimating fatigue states were extracted by calculating feature importance using LightGBM. The analysis revealed that physical fatigue strongly depends on recent heart rate indices, whereas mental fatigue strongly depends on long-term EDA indices, confirming differences in physiological characteristics according to the kind of fatigue.
The constructed classification model showed high accuracy (Accuracy: 0.634) against measured values, demonstrating that this method is effective as a fundamental technology for a practical Fatigue Risk Management System (FRMS).
This is a preview of subscription content, log in via an