Behavioral Detection and Prevention of Cheating during Online Examination using Deep Learning Approach

Abstract

With the expansion of new technologies over the last years learning has grown and universities are utilizing it in offering online exams. Cheating during has also gone up regardless of the technologies or means which universities are using. The study addressed the issues that are experienced during online evaluation of student taking exams, in universities. Currently many student engage in exam malpractice through copying during online exam. To be able to determine the behavioral metric data was downloaded from the free data repository. The data was processed, validated, trained and evaluated. Quantative research methods was used in this research. By analyzing distinct behavioral patterns and strategies employed by cheating students, the research provides valuable insights into the motivations and factors that drive such behavior. The study also identifies significant visual features present in images that indicate instances of cheating, which enhance the performance of deep learning models. Various deep learning models, including Dense Net, Mobile Net, ResNets, and Convolutional Neural Networks (CNN), are developed and evaluated for detecting and classifying cheating behavior during online examinations. The evaluation results show that the Mobile Net model achieved the highest test accuracy of 93.4%, outperforming the other models. It demonstrated strong predictive ability, accurate classification, and efficient computation time. Additionally, the identification of significant visual features and the development of deep learning models tailored for cheating detection contribute to the field of automated cheating detection, providing a foundation for future research. However, certain limitations should be acknowledged. The performance of the deep learning models may be influenced by the quality and diversity of the training dataset, and further investigation is needed to determine their effectiveness in detecting evolving cheating strategies. Based on the evaluation findings and identified limitations, several recommendations are proposed. Firstly, improving the quality and diversity of the training dataset through data collection was recommended to enhance the performance of deep learning models. Continuous model training is essential to adapt to emerging cheating strategies, requiring regular incorporation of new instances of cheating behaviors into the training dataset. Further exploration and refinement of significant visual features can enhance model accuracy through feature engineering techniques. Ensemble methods, such as model averaging or stacking, should be considered to improve overall model performance. Collaboration among researchers, educators, and policymakers from different educational contexts can facilitate cross-context evaluation and provide insights into the generalizability of the models. The findings of the research can be used by policy maker when making decision patterning online exams to ensure there is credibility of the online exams. The findings also forms the bases of academia future research to improve on this research. Ethical considerations, including privacy concerns and fairness in the detection process, should be addressed transparently. Lastly, educational institutions should prioritize creating awareness and fostering a culture of academic integrity through comprehensive guidelines and student education