Privacy-Preserving Region-Level Classroom Engagement Analytics for Teacher Reflection

Zhiyuan Xiang, Qiao Lin, Xueyan Jia, Shengkai Zheng
Proceedings of the Impactful and Responsible AI Systems for Education Workshop, PMLR 339:160-169, 2026.

Abstract

Monitoring classroom-wide students’ visible behavioral engagement is important for teacher reflection and instructional adjustment, yet video-based classroom analytics raise substantial privacy concerns when students may be identifiable from visual data. To address this tension, this paper presents a teacher-facing, region-level classroom engagement analytics framework that converts classroom video into temporal and spatial summaries while reducing reliance on identity-sensitive visual information. The framework combines stochastic head obfuscation, adversarial identity decoupling through a gradient reversal layer, and differentially private release of aggregated engagement summaries through Laplace noise injection. Using 21 short-duration classroom video segments from the Dataset for Classroom Group Engagement Recognition and a 13-minute extended instructional sequence, we examine what interpretable temporal and spatial signals the system can generate for classroom reflection across short-duration and extended analyses. We also conduct a professional evaluation with four middle school teachers to examine whether these outputs are understandable and aligned with teachers’ professional observations. The generated outputs and teacher evaluation suggest that region-level timelines and heatmaps can help teachers review moments and classroom areas that warrant contextual interpretation.

Cite this Paper

BibTeX
@InProceedings{pmlr-v339-xiang26a, title = {Privacy-Preserving Region-Level Classroom Engagement Analytics for Teacher Reflection}, author = {Xiang, Zhiyuan and Lin, Qiao and Jia, Xueyan and Zheng, Shengkai}, booktitle = {Proceedings of the Impactful and Responsible AI Systems for Education Workshop}, pages = {160--169}, year = {2026}, editor = {Basu Mallick, Debshila and Woodhead, Simon and Wang, Zichao and Ananda, Muktha and Burstein, Jill and Murphy, April}, volume = {339}, series = {Proceedings of Machine Learning Research}, month = {28 Jun}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v339/main/assets/xiang26a/xiang26a.pdf}, url = {https://proceedings.mlr.press/v339/xiang26a.html}, abstract = {Monitoring classroom-wide students’ visible behavioral engagement is important for teacher reflection and instructional adjustment, yet video-based classroom analytics raise substantial privacy concerns when students may be identifiable from visual data. To address this tension, this paper presents a teacher-facing, region-level classroom engagement analytics framework that converts classroom video into temporal and spatial summaries while reducing reliance on identity-sensitive visual information. The framework combines stochastic head obfuscation, adversarial identity decoupling through a gradient reversal layer, and differentially private release of aggregated engagement summaries through Laplace noise injection. Using 21 short-duration classroom video segments from the Dataset for Classroom Group Engagement Recognition and a 13-minute extended instructional sequence, we examine what interpretable temporal and spatial signals the system can generate for classroom reflection across short-duration and extended analyses. We also conduct a professional evaluation with four middle school teachers to examine whether these outputs are understandable and aligned with teachers’ professional observations. The generated outputs and teacher evaluation suggest that region-level timelines and heatmaps can help teachers review moments and classroom areas that warrant contextual interpretation.} }
Endnote
%0 Conference Paper %T Privacy-Preserving Region-Level Classroom Engagement Analytics for Teacher Reflection %A Zhiyuan Xiang %A Qiao Lin %A Xueyan Jia %A Shengkai Zheng %B Proceedings of the Impactful and Responsible AI Systems for Education Workshop %C Proceedings of Machine Learning Research %D 2026 %E Debshila Basu Mallick %E Simon Woodhead %E Zichao Wang %E Muktha Ananda %E Jill Burstein %E April Murphy %F pmlr-v339-xiang26a %I PMLR %P 160--169 %U https://proceedings.mlr.press/v339/xiang26a.html %V 339 %X Monitoring classroom-wide students’ visible behavioral engagement is important for teacher reflection and instructional adjustment, yet video-based classroom analytics raise substantial privacy concerns when students may be identifiable from visual data. To address this tension, this paper presents a teacher-facing, region-level classroom engagement analytics framework that converts classroom video into temporal and spatial summaries while reducing reliance on identity-sensitive visual information. The framework combines stochastic head obfuscation, adversarial identity decoupling through a gradient reversal layer, and differentially private release of aggregated engagement summaries through Laplace noise injection. Using 21 short-duration classroom video segments from the Dataset for Classroom Group Engagement Recognition and a 13-minute extended instructional sequence, we examine what interpretable temporal and spatial signals the system can generate for classroom reflection across short-duration and extended analyses. We also conduct a professional evaluation with four middle school teachers to examine whether these outputs are understandable and aligned with teachers’ professional observations. The generated outputs and teacher evaluation suggest that region-level timelines and heatmaps can help teachers review moments and classroom areas that warrant contextual interpretation.
APA
Xiang, Z., Lin, Q., Jia, X. & Zheng, S.. (2026). Privacy-Preserving Region-Level Classroom Engagement Analytics for Teacher Reflection. Proceedings of the Impactful and Responsible AI Systems for Education Workshop, in Proceedings of Machine Learning Research 339:160-169 Available from https://proceedings.mlr.press/v339/xiang26a.html.

Related Material