Cheating Detection in Examinations Using YOLOv11: A Comprehensive Real-Time Object Detection Framework for Enhancing Academic Integrity

Academic integrity in examinations is crucial forensuring fair and credible assessments, yet cheating remainsa pervasive challenge in educational institutions. Traditionalinvigilation methods, which rely on human proctors, are ofteninadequate due to scalability issues and human error. Thispaper proposes a novel real-time cheating detection systemusing YOLOv11, a state-of-the-art object detection model, tomonitor exam environments. Our system identifies six distinctbehaviors: examiner, leaning to copy, looking around, normalsitting, talking and cheating, and walking. We trained YOLOv11on a custom dataset of 5,000 exam images, achieving a meanAverage Precision (mAP) of 0.991 at a confidence threshold of0.5. The system’s performance was comprehensively evaluatedusing training and validation losses, precision-recall curves, F1-score curves, confusion matrices, and real-time detection resultsat 30 frames per second (FPS). Detection results on real examimages demonstrate high confidence in identifying key behaviors,with scores ranging from 0.9 to 1.0 for most classes. The proposedsystem offers a scalable, non-intrusive, and efficient solution forenhancing academic integrity, making it suitable for large-scaleexam settings. Future improvements include addressing classimbalance, incorporating temporal analysis, and deploying thesystem in diverse real-world environments.