- Download 14
- File Size 452.28 KB
- File Count 1
- Create Date 18/06/2025
- Last Updated 18/06/2025
Adaptive Deep Learning Approach for Monitoring Student Engagement in Virtual Classrooms
M. Navya, M. Dinesh, Y. Roshitha, K. Akhila
Mr. T. Thirupatirao Assistant Professor Department of Computer Science, Lendi Institute of Engineering and technology (JNTUGV), Vizianagaram,Andhra Pradesh,India
navyamanda8@gmail.com,dineshmahanthi7@gmail.com,roshithayerra@gmail.com,akhilak@gmail.com
Abstract — The widespread adoption of online education, especially during the COVID-19 pandemic, has brought new challenges in maintaining student engagement in virtual classrooms. Unlike traditional settings, remote learning makes it difficult for educators to gauge student attentiveness in real time. To tackle this issue, the project introduces a smart and automated engagement detection system based on ensemble deep learning methods. This system is built using a combination of machine learning models designed to analyse facial behaviour data obtained from the DAiSEE dataset. To capture detailed facial expressions and movement patterns, the Open Face toolkit is utilized for feature extraction. These features reflect various engagement states, such as boredom, confusion, and interest, providing a strong foundation for behavioural analysis. To manage the high- dimensional nature of the extracted data, Singular Value Decomposition (SVD) is employed, ensuring improved performance and reduced computational complexity. The refined data is processed through a mix of 1D Convolutional Neural Networks (1D CNN) and 1D Residual Networks (1D ResNet), which are well-suited for analysing sequential patterns in time-series data. Additionally, MobileNetV2 is incorporated into the model ensemble to take advantage of its efficiency and lightweight architecture. The system aggregates predictions from all individual models using a soft voting mechanism, enhancing the overall prediction stability and reliability. This architecture allows for robust real-time detection of engagement across different users and environments, even when facial expressions are subtle or partially obscured. A web-based interface is developed to visualize engagement status instantly, enabling integration into existing e-learning platforms. This not only supports continuous monitoring but also offers teachers actionable insights to adjust teaching methods dynamically based on student responsiveness. By enabling proactive decision-making and fostering an interactive digital environment, this approach has the potential to greatly improve the quality of virtual education. It represents a scalable and data-driven solution to one of the most pressing challenges in remote learning.
Keywords : Student Engagement, Deep Learning, Ensemble Learning, 1D CNN, 1D ResNet, MobileNetV2, Open Face, DAiSEE, Real-Time Monitoring, Educational AI, Virtual Classrooms.