Cheating Detection in Online Exams Using Transfer Learning

This has never been truer, especially with the fast growth of online education due to the pandemic around COVID19, posing a challenge for the electronic examinations‘ integrity for educational organizations. To this end, novel solutions leveraging state of the art deep learning techniques, including transfer learning models, have been launched recently. This study proposes to deploy a transfer learning framework to identify and classify cheating attempts in virtual assessments. This model analyzes candidates’ behavior and movement in real-time, as well as detecting unauthorized individuals in proximity, attempts to access mobile devices and potential aids for cheating, such as written notes. The base architecture is built off of MobileNetV2 trained on a dataset we call OEP, which we generated by screenshotting videos. Experimental results yield detection accuracy of 98.30%, superior to prior state-of-the-art models. Moreover, an ablation study based on the use of various pretrained models including InceptionV3, VGG-16, VGG-19, ResNet50 and AlexNet was performed to validate the performance of our approach. Below are the results of the evaluation, which highlight the accomplishments of our model in accurately identifying cheating during online tests, and prevent it before they happen.