Multi-Algorithm Classification of Teaching Effectiveness in Physical Education Using Motion Sensor and Engagement Data
Multi-Algorithm Classification of Teaching Effectiveness in Physical Education Using Motion Sensor and Engagement Data

Summary/Abstract: Teaching effectiveness in physical education (PE) has long been assessed through subjective observation, limiting consistency and scalability. This study introduces a data-driven framework that leverages multimodal inputs, motion sensor data from wrist-worn IMUs, student engagement metrics from classroom cameras, and textual feedback to classify instructional quality. Expert evaluators labelled 45 PE sessions as High, Medium, or Low effectiveness, forming the basis for supervised learning. Three machine learning models: Random Forest, Support Vector Machine (SVM), and XGBoost were trained on the integrated dataset. All models achieved strong classification performance, with XGBoost yielding the highest accuracy (95.1%) and consistently high ROC-AUC scores (0.99), indicating excellent discriminative power. Feature importance analysis revealed that instructor acceleration metrics were the most influential predictors of teaching quality. These findings highlight the potential of objective, multimodal analytics to enhance pedagogical evaluation and inform professional development in PE contexts.

• Details
• Contents