Predicting Students Performance Using Eye-Gaze Features in an Embodied Learning Environment

Advances in eye-gaze tracking technologies have presented the possibility of combining gaze features into students learning profiles. Building students profiles that support gaze information holds meaningful promise for embodied-based learning contexts. This paper examines the potential of developing a predictive model that incorporates student eye-gaze data to dynamically predict their learning performance during embodied-based activity using different machine learning algorithms. We have conducted an in-situ study where 110 students, aged 8-9 years, solved a tangible-based activity for learning human body anatomy. The study investigated the potential of incorporating multimodal data by comparing an eye-gaze model with a baseline model that does not include participants’ eye-gaze data. The analyses of the generated dataset indicate that including eye-gaze features significantly increased the accuracy of the Support Vector in predicting student performance by 95%. The findings suggest that combining gaze features with learning traces and behavior attributes may support modeling student learning profiles and building an adaptive embodied learning environment.