Monitoring Student Listening in Classroom Sessions: Applying Random Forest to Detect Low Listening Engagement

Abstract

This study investigates the problem of students’ lack of listening, or low listening engagement, during classroom sessions and proposes a machine-learning-based method for detecting such disengagement automatically. Drawing from existing literature on listening difficulties, attention-monitoring systems, and educational data mining, we design an approach that leverages webcam-derived behavioural features alongside classroom metadata. Key features extracted include eye-aspect ratio, gaze direction, head-pose estimation, facial-action units, and basic interaction logs captured at regular intervals.Using these multimodal inputs, we implement a Random Forest classifier to categorize each 30-second time-window of student behavior as either listening or not-listening. Experiments conducted on a custom classroom dataset demonstrate promising classification accuracy, indicating the model’s effectiveness in identifying inattentive listening patterns. The paper further presents sample results, error analysis, and a feature-importance overview.We also discuss important ethical considerations, including privacy, consent, and potential bias, along with the limitations of the current approach such as dataset size, environmental noise, and generalizability. Finally, we outline recommendations for practical deployment and future enhancements, such as real-time feedback systems and the integration of additional behavioural sensors.