Evaluation of user/operator stress using eye-tracking with machine learning algorithms

Abstract

Machine Learning (ML) is an ever-growing field that seen a bloom in recent years since the emergence of ChatGPT. Multiple studies had been done on the impact of ML on interpreting human behaviours. A lot of human factors could affect a person’s work performance, including stress, cognitive workload, fatigue, attention span, and anxiety. In this study, a stress evaluation model will be developed from eye tracking data to evaluate the stress levels of Vessel Traffic Operators. Eye trackers record multiple metrics from eye movements such as: Pupil Diameter (PD), eye openness, eye positions, fixation points and movement time. By using data collected while inducing different stress levels to participants with Stroop task, a proposed set of features will be extracted and tested on 6 chosen ML algorithms. Our study has shown that SVM performs well with features set with PD, fixation, and saccade. The proposed algorithms achieved classification accuracy of 86.24% with subject-independent training, which outperformed the best methods from other studies. Hypothesis testing was done to further prove the significance of the proposed algorithms in achieving higher classification accuracy with 95% confidence interval.