Deep Learning-based Student Engagement Detection using CNN, Mobile Net, ResNet50, VGG16

In large classrooms, tutors or student-driven learning systems may manage or summarize what is happening more easily via the detection and measurement of student attention. Today, numerous institutions can create functional systems for aiding teachers or tutoring systems alike. Deep Learning has advanced machine learning that has proved effective in automating pupil engagement tracking. The article discusses a possible solution to the problem of procuring large volumes of labeled data by suggesting a transfer learning technique for classroom-based student behavior recognition. Using several CNN architectures, MobileNet, custom CNN, ResNet50, and VGG16, this paper proposes a deep learning method for detecting student engagement. 5000 samples, equally divided between the "Engaged" and "Not Engaged" classes, were created from a lightweight dataset of 1000 images. To overcome small sample sizes and enhance generalization, we employed cross-validation and transfer learning. While ResNet50 obtained the highest raw accuracy but displayed overfitting, MobileNet provided the best trade-off between accuracy (92.5%) and efficiency among the tested models. Our results lend credence to the viability of automated classroom engagement monitoring through deep learning.