Student Engagement Detection in Classroom using Deep CNN-based Learning Approach

In a large classroom, tutors or self-directed learning systems can more easily manage or summarize the situation by detecting and measuring students' attention levels. Today, many institutions are capable of developing effective systems for assisting teachers or tutoring systems. Deep learning is a sophisticated technology in machine learning. This article suggests a transfer learning technique that may be used to classify student activities in the classroom using a limited dataset. Proposed contribution includes creating a lightweight dataset with a small number of images for each category, which can be used for classification work. In comparison to other studies, the experiments yielded satisfactory results for action recognition with high accuracy. In this work, a system that can automatically detect threats. And also want to the classification of student engagement from images using create a CNN and Mobile Net with multiple layers namely convolution, pooling, flattening, and full connection. This model achieved an overall accuracy of 93% in detecting student engagement. Proposed model is also evaluated on a separate test set that included unseen images and achieved comparable results.