Multi Modal Deep Learning Framework for Malpractice Detection in Virtual Exams Using YOLOV5

The growth of remote exams has changed education, becoming more flexible and cheaper, especially amidst global disruption like the COVID 19 pandemic. But along with this evolution, there are severe challenges that exam integrity must now face with cheating becoming increasingly more advanced as students employ misleading maneuvers and illegitimate use of external resources. To address these issues, this work investigates sophisticated techniques to detect and prevent cheating in online exams. Through motion-based frame extraction and deep learning models, namely YOLOv5, the work successfully detects and classifies four typical cheating behaviors with high precision. A real-world cheating behavior dataset was constructed, and extensive experiments proved that the proposed method outperformed state of the art methods. The specialized dataset of actual cheating actions was established, and large-scale experiments proved that the proposed method outperforms the current detection methods