<![CDATA[Student engagement is an important indicator for evaluating the quality of the learning process; however, its measurement in conventional classrooms still largely relies on subjective and labor-intensive manual observation. This study aims to establish a reproducible baseline for student engagement detection based on visual behavioral indicators using the YOLOv8 model on a public dataset. The working dataset was constructed from two subsets of the Student Class Behavior (SCB) Dataset and restructured into five behavioral classes: hand_raising, reading, writing, bowing_head, and turn_head, resulting in 9,274 image-label pairs split into 6,491 training, 1,854 validation, and 929 test samples. The experiment used YOLOv8n with an image size of 416, a batch size of 8, and 50 effective epochs in Google Colab. Performance was evaluated using precision, recall, mAP@0.5, and mAP@0.5:0.95. The results show that the model achieved a precision of 0.4429, a recall of 0.5393, mAP@0.5 of 0.4630, and mAP@0.5:0.95 of 0.3211. The best class-level performance was observed for writing (AP 0.635) and hand_raising (AP 0.597), while bowing_head (AP 0.288) and turn_head (AP 0.332) remained comparatively weak. These findings indicate that YOLOv8n is feasible as a reproducible baseline for visual student behavior detection, although annotation refinement, comparative experiments, and architectural optimization are still required to strengthen the scientific contribution and the feasibility of real-world classroom deployment]]>
Copyrights © 2026
