Related Experiment Video
Updated: Jun 23, 2026
08:25
Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
Published on: May 7, 2019
Classifying mental stress from eye tracking data: deep learning approaches for out-of-the-lab conditions
Maike Laut1, Eva Dorschky2, Robert Richer2
1Machine Learning and Data Analytics Lab, Department Artificial Intelligence in Biomedical Engineering (AIBE), Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Erlangen, Germany. maike.stoeve@fau.de.
Scientific Reports
|June 20, 2026
Summary
Unimodal eye-tracking data can detect stress without specific tasks or multiple sensors. However, reliable stress detection using eye-tracking depends heavily on data quality and signal processing methods.
Area of Science:
- Psychophysiology
- Human-Computer Interaction
- Biomedical Engineering
Background:
- Stress detection often relies on multimodal sensors or controlled lab settings, limiting real-world application.
- Existing eye-tracking stress detection methods frequently use task-specific features, hindering generalizability.
- Scalability challenges persist for stress detection in uncontrolled, dynamic environments.
Purpose of the Study:
- To investigate the efficacy of unimodal eye-tracking time-series data for task-agnostic stress detection.
- To evaluate stress classification performance in both controlled and less controlled environments using eye-tracking signals.
- To determine if eye-tracking data alone can capture stress-related physiological responses.
Main Methods:
- Analysis of pupil diameter and gaze behavior time-series data from two distinct datasets: a virtual reality goalkeeper task and a virtual job interview.
- Stress classification using unimodal eye-tracking data, evaluating performance across different recording conditions and signal quality.
- Comparison of stress detection capabilities between a controlled visuomotor task and a less controlled interview setting.
Main Results:
- Unimodal eye-tracking signals contain informative patterns related to stress-associated autonomic and oculomotor responses.
- High performance (up to [Formula: see text] macro-averaged F1-score) was achieved under favorable recording conditions.
- Significant performance variation across datasets highlights the critical impact of data quality, calibration, and task design on stress detection accuracy.
Conclusions:
- Unimodal eye tracking shows potential as a less burdensome alternative to complex multimodal systems for stress detection.
- Reliable stress detection using eye-tracking is fundamentally dependent on the interplay between data characteristics, signal representation, and modeling approaches.
- Further research is needed to optimize eye-tracking-based stress detection for diverse and uncontrolled real-world scenarios.
