A Deep Visual Approach to Student Engagement Analysis Using Affective and Behavioral Cues

Abstract

Assessing student engagement in educational environments is essential to support adaptive teaching strategies and enhance learning outcomes. This study presents a deep learning-based approach for automatically predicting student engagement, leveraging both behavioral and emotional cues. The proposed method integrates features derived from facial emotion recognition and head pose estimation, capturing a comprehensive representation of student affect and attention. Using the Student Engagement Dataset, a multi-layer neural network was trained to classify engagement states based on these multimodal inputs. The proposed framework achieves an accuracy of 88% on unseen validation data, demonstrating strong effectiveness in distinguishing between engaged and disengaged students. In addition, explainability analysis highlights the importance of neutral facial expressions and head orientation as key indicators of engagement, supporting the interpretability and practical relevance of the proposed approach for real-world educational environments.