Behavioral-based Real-Time Cheating Detection in Academic Exams using Deep Learning Techniques

Recently, with the development of Artificial Intelligence (AI), the use of automated evaluation in education has increased. Maintaining academic integrity is one of the most challenging aspects of higher education. Cheating is rampant in academic examinations and other forms of educational assessment. The vast majority of students believe that it is unethical to tolerate cheating; therefore, it is vital to devote a significant amount of effort to identifying and avoiding instances of cheating. Examining the student’s behavior is one way to determine whether they are engaged in cheating or not. This paper proposes a deep learning-based cheating detection system that can identify instances of students engaging in dishonest behavior. A YOLOv7 model is trained on a custom dataset collected from various resources. The dataset comprises two classes, i.e., cheating and not cheating, and 2565 images. Evaluation criteria like precision, F1 score, recall, and mAP (mean average precision) are used to validate the performance of the proposed model. The proposed model shows promising performance in categorizing the student’s visible actions into cheating or not cheating and achieved an overall [email protected] of 0.719. Overall, the proposed method can be utilized to reduce the error rate associated with human monitoring by alerting the proper authorities whenever unusual behavior is observed during academic tests.