Loading content...
Teachers are wary of letting students take tests
online because of the rising incidence of academic dishonesty,
making fraud detection in online assessment an increasingly
pressing issue. This became more of a problem during the
coronavirus epidemic, when there was a pressing need to
swiftly convert in-person classes, seminars, labs, and
assessment assignments into online alternatives. Many
students also find the altered activities more difficult, adding
fuel to the fire of student-instructor friction about the
prospect of fraudulent collaboration on independently
administered assessments. This study employs AI techniques
to overcome these obstacles; first, it uses data preprocessing
and SMOTE to fix class imbalance; then, it uses feature
engineering to make the model better at discriminating. In
addition, social media posts are analysed using natural
language processing (NLP) techniques to gain a better
understanding of trends of online exam cheating. With this
goal in mind, it create a new hybrid architecture called
"FastText CNN with LSTM." This architecture combines
CNN's global feature extraction capabilities with fastText
embeddings' semantic text representation and LSTM's
sequential dependency learning capabilities. This method
outperforms previous models, providing a solid foundation
for detecting online assessment fraud, with an accuracy rate
of 94.50%
Nowadays, most Internet transactions are
conducted using wireless mobile terminals, that can
authenticate users using their passwords, fingerprints,
noises, and photos. In order to commit fraud, the scammer
can gather user data, including ID, password, age,
occupation, and other details, and log in to several trading
systems as an actual user. The fast evolution of information
technology has made this type of fraudulent activity
widespread, and it does enormous harm to consumers,
companies, and society at large[1]. Further, conventional
information security measures will not stop online transaction theft after these personal details have been
obtained. Internet platforms must consequently develop
anti-fraud mechanisms to safeguard these new services.
Since most popular online fraud detection systems use
labelled data to build robust models, this data shortage
could make current fraud detection algorithms useless[2].
Labelled data plays a big role in identifying service fraud
or benign identities, but both parties need to collect enough
business data to make a determination. Consequently, the
newly-launched online services' fraud detection system
experiences the infamous cold-start problem. Despite the
scarcity of labelled data, millions of user behavioral records
are continually amassed across all types of internet sites.
They could be seen as unlabeled secondary data sources
that could encourage us to use them to enhance fraud
detection performance. The rise of Internet-based financial services
reliant on cutting-edge innovation like big data has been
steadily gaining prominence, due to the expansion of the
web. Unfortunately, people and companies have also fallen
victim to related forms of online financial fraud, such as
those involving credit cards, bank statements, insurance,
etc. Online banking has made many people's lives easier,
but unfortunately, there is a dark side to this convenience:
the prevalence of transaction fraud[3]. The field of study
devoted to developing models and technology for the
identification of fraudulent transactions has been growing
in the realm of online finance and electronic transactions.
One of the biggest problems with financial services is the
amount of money that can be lost due to fraudulent activity,
especially when using credit cards. Finding fraudulent
transactions quickly is critical for a number of reasons,
including lowering financial losses, improving customer
service, and protecting financial institutions' reputations.
Finding a solution to this problem will require collaboration
between academics and financial institutions to test and
refine learning frameworks for reducing false positives in
fraud detection systems and generalizing information from