CSBAM-MobileNet: a lightweight attention-enhanced deep learning model for student attentiveness classification

Emotions significantly influence behaviour and decision-making, making the assessment of student attentiveness a critical factor in optimizing learning outcomes. This paper introduces a vision-based deep learning framework designed to evaluate student attentiveness through facial expression analysis. Leveraging a Customized lightweight MobileNet architecture enhanced with Channel and Spatial Attention Mechanisms (CSBAM), the model efficiently captures subtle facial cues relevant to attentiveness, such as eye closure, gaze direction, and mouth movements. A custom classroom expression dataset was curated for training, enabling the model to achieve a high accuracy of 91.2%. On more challenging subsets exhibiting varied lighting, occlusion, and head pose variations, the model maintained robust performance with 84.2% accuracy. Evaluation across multiple public benchmark datasets further demonstrated the model’s adaptability, achieving accuracies of 92.5% on DAiSEE, 89.6% on RAF-DB, and 87.4% on FER-2013.This framework offers real-time insights into student engagement, providing a foundation for adaptive and personalized teaching strategies. Future research aims to integrate additional behavioural and physiological cues, such as body posture and voice features, to enable a more comprehensive and multi-modal assessment of student attentiveness. The proposed approach holds promise for enhancing educational interventions and fostering active learning environments.