EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-SA 4.0

Student Behavior Detection in Classroom Environments Using Deep Learning Models

Thi-Nguyen Nguyen, Dinh-Thai Kim, Anh-Phuong Pham

EAI Endorsed Transactions on AI and Robotics · 07.04.2026 · 5

DOI: 10.4108/airo.11141 ↗

Kurz gefasst

Fachartikel in der Zeitschrift EAI Endorsed Transactions on AI and Robotics, online erschienen am 07.04.2026. Verfasst von Thi-Nguyen Nguyen und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-SA 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Observing student behavior in classroom environments provides valuable insights into attention and engagement. However, manual monitoring is labor-intensive and often inconsistent, particularly in large lecture settings. This study proposes a deep learning–based framework for automatic student behavior analysis in real classroom scenarios. A dedicated dataset was constructed, consisting of 3,373 images annotated with five attention-related behaviors—“Focused”, “Raising Hand”, “Distracted”, “Sleeping”, and “Using Phone”—totaling 9,659 labeled instances. The dataset captures diverse real-world conditions, including variations in classroom layout, camera viewpoints, and occlusion. We systematically evaluated stateof-the-art object detection models, including YOLOv8–YOLOv12 and the Real-Time Detection Transformer (RT-DETR). Experimental results show that YOLOv8m achieved the highest localization accuracy (mAP@0.5 = 0.920), YOLOv11s/m demonstrated the best overall performance (mAP@0.5:0.95 = 0.726), and RT-DETRX achieved the highest F1-score (0.886). Notably, larger model size does not necessarily translate into better performance. In addition to accuracy, inference speed was evaluated to assess real-time applicability. Lightweight models such as YOLOv11s achieved a favorable balance between performance and efficiency, enabling real-time processing on resource-constrained hardware. Furthermore, YOLOv11s—with high Precision (0.890) and only 9.4M parameters—was integrated with ByteTrackV2 to perform behavior tracking and temporal analysis in classroom environments. This enables the generation of behavior distribution charts that provide interpretable insights into student engagement over time. These findings demonstrate the potential of automated behavior recognition systems for classroom analytics and data-driven teaching improvement.

Zitieren

APA

Nguyen, T., Kim, D., & Pham, A. (2026). Student Behavior Detection in Classroom Environments Using Deep Learning Models. EAI Endorsed Transactions on AI and Robotics, 5. https://doi.org/10.4108/airo.11141

Vancouver

Nguyen T, Kim D, Pham A. Student Behavior Detection in Classroom Environments Using Deep Learning Models. EAI Endorsed Transactions on AI and Robotics. 2026;5. doi:10.4108/airo.11141

BibTeX
@article{Nguyen2026,
  author = {Nguyen, Thi-Nguyen and Kim, Dinh-Thai and Pham, Anh-Phuong},
  title = {Student Behavior Detection in Classroom Environments Using Deep Learning Models},
  journal = {EAI Endorsed Transactions on AI and Robotics},
  year = {2026},
  volume = {5},
  doi = {10.4108/airo.11141}
}
Kostenlos starten →