Abstract
The modernization of K-12 educational environments has led to the widespread adoption of flexible smart classrooms designed to foster active learning, peer interaction, and dynamic group collaboration. However, evaluating how physical spatial configurations influence collaborative learning dynamics remains a key challenge for educators and researchers. This study presents a comprehensive Multimodal Learning Analytics (MMLA) framework designed to unobtrusively measure and analyze spatial collaboration and peer interaction within flexible K-12 smart classrooms. Utilizing an integrated sensor architecture comprising ultra-wideband location tracking, ceiling-mounted computer vision cameras, and directional microphone arrays, we captured fine-grained spatial trajectories, body posture, and vocal turn-taking patterns from 120 middle school students across 16 collaborative learning sessions. Machine learning classification models and spatial graph analytics were deployed to map physical positioning against collaborative engagement levels validated by human coders. The results demonstrate that high-performing collaborative groups exhibit higher spatial mobility, balanced vocal distribution, and optimal physical proximity clusters between 0.8 and 1.2 meters. Furthermore, spatial heatmaps identified critical furniture layout configurations that either facilitated or restricted non-verbal peer interactions. This research advances MMLA methodology by providing a scalable computational approach to understanding spatial pedagogy, offering actionable insights for designing active learning spaces and informing automated real-time instructional scaffolding in smart K-12 environments.
