Fast and accurate cheat detection using deep learning algorithms

Abstract

The rapid expansion of online education, accelerated by the global health crisis of 2020, has introduced significant challenges in maintaining academic integrity due to the absence of physical supervision during remote examinations. As digital learning becomes a permanent component of modern education, ensuring fairness and credibility in online assessments has become a critical concern for educational institutions. This study proposes an intelligent deep learning (DL)–based framework for detecting non-compliant behaviors during online examinations using standard webcam video streams. The proposed system integrates real-time video monitoring with automated behavioral analysis by combining object detection and classification models. In particular, you only look once version 5 (YOLOv5) is employed for efficient facial and object detection, while a convolutional neural network (CNN) is used to classify cheating and non-cheating behaviors from extracted visual features. Experimental results demonstrate that the integrated YOLOv5–CNN architecture achieves high detection accuracy and low inference latency, making it suitable for real-time and scalable deployment in online proctoring systems. By enabling objective and automated monitoring, the proposed framework contributes to strengthening fairness, transparency, and trust in digital assessment environments, thereby supporting the long-term sustainability of online education.