Published Paper Details:

Online examinations have gained significant prominence due to their flexibility and cost efficiency, especially during extraordinary situations such as the COVID-19 pandemic. Nevertheless, the absence of direct human supervision introduces serious challenges related to academic dishonesty. This paper focuses on detecting abnormal behaviors during online examinations to uphold academic integrity and analyzes the performance of various deep learning models, including DenseNet121, YOLOv5, and its advanced variant YOLOv8, for identifying typical cheating activities such as unusual head movements and the usage of electronic devices. A custom dataset comprising four categories of cheating behaviors was developed, along with an additional dataset obtained from Roboflow containing two classes: 'No Head Movement' and 'Head Movement. Evaluation metrics such as accuracy, precision, recall, and mean Average Precision (mAP) were utilized to comprehensively assess the performance of the proposed models. DenseNet121 achieved an accuracy of 85%, whereas YOLOv5 recorded over 97% in precision, recall, and mAP, demonstrating its effectiveness for real-time object detection. Furthermore, YOLOv8 exhibited superior performance, achieving an accuracy of 98.42%, precision of 97.95%, recall of 98.18%, F1-score of 98.06%, and mAP of 98.71%. These results indicate that YOLOv8 provides enhanced detection accuracy, improved localization capability, and robust real-time monitoring of candidate activities, making it highly suitable for intelligent online exam proctoring applications. The proposed framework provides a reliable, efficient, and scalable solution for automated online examination monitoring, significantly reducing the need for manual supervision while enhancing the fairness, security, and integrity of remote assessment environments