2026 ASEE Annual Conference & Exposition

Understanding classroom engagement is essential for improving teaching effectiveness and promoting inclusive learning environments. However, traditional human observation methods often face limitations, especially in large or culturally diverse classrooms where students may be reluctant to ask questions or express emotions openly. To address these challenges, this work presents a vision-based artificial intelligence system developed to monitor and analyze student engagement in real time.
A convolutional neural network was fine-tuned using locally collected and incrementally expanded training data to better capture behavior patterns across different classroom contexts. The system provides three types of feedback: (i) live engagement visualization during lectures, (ii) periodic summary reports at defined intervals, and (iii) post-class analytics that categorize engagement states such as interested, disengaged, confused, tired, and talking. After data augmentation, approximately 15,000 labeled frames were used for retraining, allowing the prototype to process classroom video at about 15 frames per second on standard laptop hardware, achieving near–real-time operation without specialized equipment.
Preliminary evaluations were conducted across classrooms in Kuwait and the United States. Accuracy on a held-out test set reached approximately 87%. Cross-validation with in-class surveys showed an agreement of approximately 82%. Cross-cultural variations, such as partial facial occlusion (e.g., head coverings) and variations in expressive behaviors, modestly influenced the model’s sensitivity to engagement cues.
Ongoing work focuses on expanding dataset diversity, improving temporal analysis methods, and conducting large-scale evaluations to assess robustness, fairness, and pedagogical value. Results indicate the feasibility of a low-overhead, privacy-conscious AI framework that provides educators with practical insights into classroom engagement across varied cultural and instructional settings.