| dc.description.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 | en_US |
