Alexis Torres¹, Gene Brewer²; ¹Arizona State University, ²University of California, Riverside

Sustained attention declines over time, leading to increased attention failures. Eye metrics, such as prestimulus pupil size, stimulus-evoked pupil responses, and gaze stability, have been identified as indices of task engagement and predictors of attention failures at the group level. This study evaluates the predictive power of eye metrics for impending attention failures at the trial level in single subjects using machine learning algorithms. In this study, 225 undergraduate students completed the psychomotor vigilance task. Eye metrics—including mean prestimulus pupil size and variability, mean stimulus-evoked pupil response amplitude and latency, and preparatory gaze stability—were extracted for each trial. Multiple machine learning algorithms were trained to predict and classify response times based on these metrics at both the group and subject levels. Group-level predictions showed robust accuracy, with models effectively predicting trial-level response times. However, single-subject trial predictions exhibited lower sensitivity, particularly for slower response classifications. These findings suggest that while eye metrics are reliable predictors at the group level, additional refinements are needed to improve their predictive power for individual performance. This highlights the potential of eye metrics as tools for understanding attention dynamics and underscores the challenges of translating group-level insights to single-subject applications.

Acknowledgements: National Defense Science and Engineering Graduate (NDSEG) Fellowship Program