Real-World Classification of Student Stress and Fatigue Using Wearable PPG Recordings

Download PDF
Open Webpage

Justin Laiti, Yu Liu, Pádraic J. Dunne, Elaine Byrne, Tingting Zhu

Published: 2026, Last Modified: 28 Mar 2026IEEE Trans. Affect. Comput. 2026EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Wearable-based affective computing offers a promising solution for monitoring and managing stress and fatigue in adolescent student populations. This can help to support student well-being without increasing reliance on mobile phones by providing on-device insights into stress and fatigue levels. To do so, the processing and classification methods must be lightweight enough to be executed in real-time on wearables. This study outlines the findings of the implementation of a student-informed wearable and mobile app, Wellby. Students tested this for one month while completing routine photoplethysmography (PPG) recordings and check-ins on their perceived levels of stress and fatigue. This study proposes a lightweight processing pipeline, intended for wearable-based deployment, while examining its classification performance on real-world student PPG data from Wellby. The pipeline performs denoising, fixed noise elimination, and peak detection to calculate time-domain heart rate variability (HRV) metrics. It was first evaluated on public datasets, the Wearable Stress and Affect Detection (WESAD) dataset and the AKTIVES dataset, achieving an area under the receiver operating characteristic curve (AUC-ROC) of up to 91.58% for stress classification on WESAD and 76.61% on AKTIVES. In the Wellby dataset, the adapted processing pipeline achieved an AUC-ROC of 77.02% for stress classification and 71.58% for fatigue classification using only time-domain HRV features. Furthermore, the inclusion of a signal quality metric and baseline well-being questionnaires improved the AUC-ROC for stress classification to 91.60% in the best performing model. These findings demonstrate the potential for wearables to implement real-time affective computing, providing timely feedback to students in real-world settings based on PPG and contextual data.