REAL-TIME STUDENT ENGAGEMENT DETECTION THROUGH FACIAL EXPRESSIONS USING CNN AND YOLO

Abstract

Understanding students’ emotional engagement is essential for optimizing teaching strategies and improving classroom learning outcomes. This study presents a real-time facial expression recognition system using Convolutional Neural Networks (CNNs) integrated with YOLOv11 to identify students' learning interest during classroom sessions. The system classifies six facial expressions—happiness, surprise, sadness, fear, anger, and disgust—into interested or uninterested categories. The model was developed using a custom dataset of 3,500 annotated student images captured in authentic classroom conditions. A prototyping approach was adopted to build and refine the system through iterative testing. The model achieved 86% classification accuracy with real-time performance averaging 24 frames per second. It was successfully deployed in live classroom settings, enabling teachers to adjust their instruction based on students' emotional feedback. Findings indicated that teachers responded positively to the system's feedback, using it to introduce interactive activities and modify delivery styles when signs of disengagement appeared. Compared to more complex models, this system offers a cost-effective and efficient solution adaptable to typical school environments. Its implementation supports the growing movement toward emotion-aware and AI-driven pedagogy. This study contributes to the body of knowledge in educational affective computing and suggests promising future directions, including the use of multimodal emotion analysis and long-term impact studies on learning engagement.