Stress Detection Using Eye Activity via Recurrent Neural Networks.

Area of Science:

• Computer Science
• Biomedical Engineering
• Psychology

Background:

• Chronic stress significantly impacts well-being and can cause physiological and psychological disorders.
• Current automatic stress detection methods often rely on intrusive physiological sensors, limiting practical, everyday use.
• Existing sensor-based methods face accuracy constraints and user discomfort.

Purpose of the Study:

• To develop a novel, non-intrusive method for automatic stress detection.
• To investigate the efficacy of using eye data (gaze direction and landmarks) for stress detection.
• To compare the performance of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks for this task.

Main Methods:

• Utilized eye gaze direction and eye landmarks captured by a standard video camera.
• Employed two types of Recurrent Neural Networks (RNNs): Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU).
• Experimented on the UBFC-Phys dataset to evaluate the proposed method.

Main Results:

• Stress detection accuracy reached 83.69% using only gaze data.
• Incorporating eye landmarks improved accuracy to 90.12%.
• Combining gaze and landmark data achieved a maximum accuracy of 90.83%.

Conclusions:

• Eye data, specifically gaze direction and landmarks, provides an effective basis for accurate, non-intrusive stress detection.
• Recurrent Neural Networks (LSTM and GRU) are suitable for analyzing temporal eye data for stress detection.
• The proposed camera-based approach offers a practical and accurate alternative to traditional physiological sensing methods.