An eye-movement-based interpretable machine learning model for detecting stress: real-time monitoring in logistics working environments

Abstract

Stress is a common mental state in the work environment. Moderate level of stress can contribute to work performance, while excessive stress can lead to serious operation failures and psychological problems. This study aimed to propose an eye-tracking-based machine learning model to detect human stress in logistics working environment to identify workers' high stress status to avoid non-technical errors caused by poor mental state. In the case study, a computer screen-based operating scenario was built to simulate the text processing work in logistics. Firstly, 43 students with logistics and manufacturing academic background participated in a text processing task and their eye-movement data were recorded. After the task, they were asked to fill out the Short Stress State Questionnaire (SSSQ), which is a valid measure of stress state. Secondly, various state-of-the-art machine learning algorithms were compared and LightGBM (Light Gradient Boosting Machine) was found to achieve the best performance with high accuracy of 92.71%, 87.60%, and 90.83% in detecting participants' stress states of task engagement, worry, distress, respectively. Thirdly, an interpretable model, SHAP (SHapley Additive exPlanation), was used to explain the main effects of eye-movement features on detection. The case study demonstrates the effectiveness of combining eye-movement data with machine learning methods for identifying subjects with high-stress level in text processing task. This finding also provides a practical and theoretical basis for applying eye movements to monitoring mental states and cognitive performance in other logistics and manufacturing fields such as warehouse, air traffic, port shipping, etc. in the future.