Abstract:
As educational assessments migrate to digital platforms, ensuring academic integrity becomes crucial. Traditional plagiarism detection systems struggle to catch instances of intelligent cheating, especially when students use advanced generative AI tools to produce responses. This thesis proposes a novel approach based on keystroke dynamics to distinguish between bonafide and assisted writing in academic contexts. A dataset was created to record the keystroke patterns of individuals during writing tasks, with and without generative AI assistance. Machine learning and pattern recognition techniques were employed at various levels of detail to determine if unusual typing behaviors can indicate academic dishonesty. The detectors were evaluated under various condition-specific and condition-agnostic scenarios, and the findings reveal noticeable differences in keystroke dynamics recorded for genuine and assisted writing. The thesis outcomes contribute to understanding the interaction between users and generative AI, with implications for enhancing the reliability of digital educational platforms.
