Adaptive AI Algorithms for Proctoring: Automating Student Behaviour Profiling and Real-Time Misconduct Detection

Abstract

Abstract This thesis presents the design and evaluation of an adaptive artificial intelligence (AI) based online proctoring system, aiming to automate student behavior profiling and enable real-time misconduct detection. Motivated by the rapid transition to online assessments during the COVID-19 pandemic, the study addresses the limitations of traditional rule-based proctoring systems, including high false-positive rates and fairness concerns. The proposed system combines rule-based anomaly detection with a novel adaptive behavior profiling module, which learns individual student behaviors during mock exams to distinguish between genuine violations and innocuous, habitual movements. Multiple modules, such as gaze tracking, head pose estimation, murmur detection, and environmental noise detection, were developed and integrated into a hybrid framework. A series of controlled mock and simulated real exams with student volunteers demonstrated the system’s effectiveness in improving precision, recall, and fairness while maintaining real-time performance. Evaluation results showed significant reductions in false positives and increased user acceptance of the adaptive profiling concept. This work contributes to the field of intelligent proctoring by proposing a scalable, modular, and behaviorally aware detection framework that balances academic integrity with student experience, laying the groundwork for future research on ethical, inclusive, and robust online examination systems.