Author Information
Ohnmar Nyunt, Kobe Institute of Computing, JapanSokout Hamidullah, Kobe Institute of Computing, Japan
Abstract
The COVID 19 pandemic and the 2021 political issue have caused significant disruption to campus based exams in Myanmar, the diaspora students unable to finish their degrees. To facilitate academic continuity, this research was to determine detailed user requirements through surveys with 51 students and 11 invigilators, and to use these findings to develop an online proctoring system that integrates lightweight multi-models to identify overt and covert cheating actions in real time. The system design includes multi-module: user authentication through one-time password (OTP) code and facial recognition; real time detection that streams camera, microphone, and system log information; object detection to flag prohibited items; voice detection for spoken responses; face persistence to ensure that only one authorized face is visible; and active window monitoring that prevents copy, paste, multi-tab and multi-app. A mixed methods pilot was conducted with 20 students and 14 invigilators who each took a 15 minute essay test. Quantitative data was collected through a post exam questionnaire using 5 point Likert constructs. The teacher group rated a high mean score (4.50) with low variance (0.77), while the students rated with a mean of 4.12 and standard deviation of 1.41. The system indicates a high and balanced accuracy metrics with F1-score of (91.41%) which confirms high reliability in both detecting cheating and preserving legitimate activity. Qualitative results emphasize the strengths of the proposed system, including cost-effectiveness and fairness with less bias. The results of this research indicate that the AI-based multi-modal proctoring system can be used as an academic integrity solution for online exams.
Comments
Powered by WP LinkPress