همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Early Prediction of Inhibitory Control Lapses from Unconstrained Gaze Dynamics

نویسندگان :

Rezvane Mirzapour¹ Farhane Mirzapour² Mehdi Delrobaei³

1- دانشگاه صنعتی خواجه نصیرالدین طوسی 2- دانشگاه صنعتی خواجه نصیرالدین طوسی 3- دانشگاه صنعتی خواجه نصیرالدین طوسی

کلمات کلیدی :

Biomechatronic Systems،Affective Computing،Cognitive Errors،Predictive Modeling،Gradient Boosting

چکیده :

Predicting human cognitive errors from early behavioral signals is vital for responsive human-computer interaction but often depends on specialized eye-trackers, limiting scalability. This study explores whether it is possible to predict inhibitory control lapses early using a standard laptop webcam. We have created a computer vision pipeline that analyzes webcam video to extract gaze coordinates, which are refined and classified into directions (left, center, right). Gaze dynamics features from the first 10-25% of a trial (150-375 ms post-stimulus) were used in an emotional Simon task. A Gradient Boosting classifier, trained via Leave-One-Participant-Out cross-validation, distinguished correct from error trials. The model predicted subsequent response errors with remarkable performance, achieving 97.0% accuracy and an F1-score of 71.0%, significantly outperforming baseline models. Findings show that early gaze data reflect inhibitory processing, providing a framework for cognitive monitoring. This supports practical applications such as personalized training and driver safety.