Online Supervision-Based System Using CNN for Detecting Fraud in Exams

Abstract

Maintaining academic integrity during exams is crucial in the rapidly evolving digital era. This study proposes a student behaviour monitoring system using deep learning technology, specifically Convolutional Neural Networks (CNN), to detect cheating in real-time. The system combines the Single Shot Multibox Detector (SSD) for face detection and the Haar Cascade algorithm for eye detection, enabling the analysis of behaviours such as focus, suspicion, and cheating during exams. Results show that the SSD model achieves over 95% accuracy for face detection under adequate lighting, while the Haar Cascade achieves around 90% accuracy for eye detection. This method effectively detects cheating with a 92% accuracy rate, provides immediate feedback to exam proctors, and stores behaviour data for further analysis. The primary contribution of this research is developing an efficient and reliable exam monitoring system that not only detects cheating in real time but also provides comprehensive behaviour data to enhance exam security and integrity. Additionally, the system's ability to store and analyze behaviour data allows continuous improvement and customization to address various cheating patterns. Although further development is needed to improve accuracy under low lighting conditions and implement advanced machine learning technologies, this method can significantly enhance exam monitoring. It offers a more comprehensive and reliable solution than previous methods, contributing to upholding academic standards in online and traditional exam settings.