AI Framework for Exam Cheating Detection Uses YOLOv8n and RexNet-150 Models

A new two-stage deep learning framework aims to improve exam cheating detection by combining object localization with behavioral analysis. The system first employs the YOLOv8n model to identify students in exam-room images, then uses a fine-tuned RexNet-150 model to classify behavior as normal or cheating. Training involved a dataset compiled from 10 independent sources containing 273,897 samples, achieving 0.95 accuracy. This approach addresses limitations of traditional human invigilation, which is often inefficient and error-prone at scale. While some AI monitoring systems already exist, many lack transparency or require complex architectures. The proposed framework integrates established technologies to enhance performance and reliability in academic integrity enforcement.