Using multimodal learning analytics as a formative assessment tool in AI-assisted physical education: a case study of Baduanjin teaching
Year: 2026
Type: article
Abstract: Introduction Traditional formative assessment in physical education (PE) often lacks objectivity and fails to capture the intricate multimodal dynamics of instructional behaviors. This study validates a Multimodal Learning Analytics (MMLA) framework to decode teaching effectiveness in a peer-coaching context using the traditional Chinese exercise, Baduanjin. Methods We analyzed synchronous data from 20 instructional dyads ( N = 20 ), combining kinematic pose estimation (MediaPipe), speech fluency metrics (Whisper), and facial emotion recognition (OpenFace). Results Results from Spearman correlations and Mann-Whitney U tests revealed a “Precision-Economy” instructional archetype: high-performing instructors were characterized by superior kinematic fidelity ( r s = .52 ) and verbal economy ( r s = .47 ), rather than sheer feedback volume or emotional intensity. Counter-intuitively, excessive corrective feedback negatively correlated with learner skill gains ( r s = − .41 ), suggesting a cognitive load interference effect. Discussion These findings challenge the “more-is-better” pedagogical assumption and demonstrate that AI-driven analytics can objectively quantify the tacit mechanisms of embodied instruction, offering a scalable tool for teacher training and developing countries seeking to modernize PE assessment, while acknowledging the necessity for further validation in open-skill sports contexts.
Source: Frontiers in Education
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: gold
Estimated APC paid: $2,353
OpenAlex ID: W7140084381