Online exams are becoming increasingly common. However, online
exam systems remain vulnerable to cheating, such as using the “inspect
element” feature or opening other tabs. This study aims to develop and
test an automated monitoring system using the Random Forest algorithm
to detect cheating in real time. This system utilizes data on
exam takers’ behavior, such as the number of page-limit violations, screenshots,
developer mode activity, and interaction logs.
The system was tested using 173 exam activity session datasets from students that
included instances of both honest and cheating behavior. To ensure the accuracy
of predictions across the entire dataset, this study applied the 5-Fold Cross
Validation. The test results showed that the Random Forest algorithm can
detect cheating with an accuracy rate of 93.68 percent. The system
recorded 7 false positives and 4 prediction errors
out of a total of 173 tests. In addition, this system has an average prediction processing time
of 5.83 milliseconds, which is below the normal limit of 2 seconds. In conclusion,
the Random Forest algorithm can be used as the primary system for detecting
cheating in online exam applications to maintain students’ academic integrity