Abstract:
Essay-based E-exams require answers to be written out at some length in an E-learning
platform. The questions require a response with multiple paragraphs and should be logical and well-
structured. These type of examinations are increasingly becoming popular in academic institutions of
higher learning based on the experience of COVID-19 pandemic. Since the exam is mainly done virtually
with reduced supervision, the risk of impersonation and stolen content from other sources increases. Due
to this, there is need to design cost effective and accurate techniques that are able to detect cheating in
an essay based E-exam. In this work we develop, train and evaluate real-time LSTM, RNN and GRU
algorithms, and then benchmark the performance of the algorithms against other state-of-the-art models
in the same study area of detecting cheating in exam in an E-learning environment. Based on a set
threshold, the models alert on possible impersonation or stolen content if the discrepancy exceeds the
threshold. The evaluation and benchmarking of the algorithms revealed that our GRU model has the
highest accuracy of 98.6% compared to other models in similar studies.
