• head pose estimation
• deep learning
• attention level
• wearables

Understanding and enhancing students' attention is critical for effective learning outcomes in educational technology. This thesis presents a novel approach to monitoring students' attention by combining triaxial magnetometer data for ground truth generation with advanced deep learning techniques. A triaxial magnetometer accurately captures the ground truth of students' attention direction, providing a reliable reference for the analysis. The thesis investigates the feasibility and effectiveness of using a deep learning model to predict whether students are focusing their attention on a specific point of interest. In particular, the proposed technique uses depth maps generated as input for training and evaluating the deep learning models. This allows us to predict students' attentional focus, contributing to the growing field of educational analytics by offering an advanced solution for real-time monitoring of student attention. During the experiments, we collected thousands of depth maps from three subjects. The test evaluations showed high levels of accuracy in assessing the students' point of focus, indicating the potential of the proposed approach to provide valuable insights into students' engagement with the learning content. The results open new lanes for further exploration at the intersection of sensor technology, deep learning, and educational psychology, with implications for designing and implementing more adaptive and personalized learning environments.