Elly Matul Imah, Riskyana Dewi Intan Puspitasari, Fadhilah Qalbi Annisa, Hasanuddin Al Habib
Academic cheating is a serious problem in education. One of the cheating that often happens is cheating during exams. This study uses surveillance camera recordings to present a novel approach for detecting cheating and suspicious activities during exams. This system allows monitoring of students during exams by observing suspicious activity that indicates cheating, such as cheating from a book, notes, or any text found on paper; talking to a person in the room; using the Internet; asking a friend a question over the phone; and using a phone. Leveraging the power of deep transfer learning architecture effectively captures intricate surveillance footage features. Additionally, the Long short-term memory (LSTM) network is utilized to model temporal dependencies in the sequence of frames, enhancing the accuracy of cheating detection. Experimental results show the efficacy of the proposed approach, achieving a commendable accuracy level of 0.964 and MCC 0.971. Furthermore, this study undertakes a comparative analysis against alternative architectures VGG-16, VGG-19, Resnet50V2, and InceptionResNetV2. The ResNet50V2-LSTM approach outperforms these models in cheating detection performance. This accomplishment underscores the potential of deep learning techniques in addressing the intricate task of detecting cheating during exams. Transfer learning with ResNet50V2 demonstrates the model's ability to learn and generalize features relevant to cheating behaviors, while the integration of LSTM effectively captures the temporal context within the video sequences. The study contributes to the broader discourse on utilizing advanced technologies to uphold the integrity of academic evaluations and promotes further research in refining cheating detection methodologies. © 2023 ACM.
Data Science Department, Universitas Negeri Surabaya, Indonesia
