Academic assessments have long relied on written exams as a primary evaluation method, but the increasing prevalence of exam malpractice undermines their effectiveness. Cheating, particularly in time-bound assessments, has become a global concern. Traditional manual invigilation remains labor-intensive and prone to human error, often failing to detect subtle dishonest behaviors such as unauthorized interactions. Existing surveillance methods lack the precision required for modern exam settings, where detecting malpractice is increasingly complex. To address the challenges, AI-driven techniques leveraging machine learning models like YOLOv8 and RetinaNet are explored for real-time detection of suspicious behaviors during offline examinations. The detection systems analyze actions such as head-rotation, note-passing, and unauthorized communication, enhancing accuracy while reducing reliance on human oversight. Training the models on datasets incorporating varied environmental conditions ensures robust detection and classification of unethical behaviors. The comparative analysis demonstrates that YOLOv8 outperforms RetinaNet in controlled environments with a precision of 0.96, whereas RetinaNet excels in dynamic environments with a precision of 0.91, offering higher precision, in their respective domains. The outcomes of the paper help to strengthen proctoring systems, minimize human biases, errors, and reinforce academic integrity by ensuring a fairer conduct of examinations.
原文链接
分享
引用
