Sensitivity Preservation and Precision of Plagiarism Detection Engines for Modified Short Programs.

Source code plagiarism presents a continual threat to the integrity and effectiveness of engineering education, as habitual cheating often has devastating impacts on students' academic and professional careers. As programming becomes an increasingly central component of first-year engineering curricula, it is essential that instructors are able to uphold academic integrity by identifying students who engage in misconduct, either through direct plagiarism or excessive peer collaboration. Instructors have an arsenal of plagiarism detection tools at their disposal, and students are keenly aware of this. Consequently, in an attempt to evade detection, students routinely make superficial modifications to plagiarized work prior to submission. Effective plagiarism detection tools attempt to mitigate the effect of these alterations, however, the extent to which precision can be maintained for heavily modified code is limited. One aim of this paper is to quantify the effect of code modification strategies on a plagiarism detection tool's ability to preserve both sensitivity to plagiarism and precision of results. This paper will introduce a novel dimensionless metric apt for the evaluation and comparison of a plagiarism detection tool's robustness to code modification. The specific context of engineering education presents additional challenges, as research in plagiarism detection methods and performance is often not applicable to short programs in dynamically typed languages which constitute typical submissions in first-year engineering coursework. This paper will analyze the performance of relevant plagiarism detection tools on short Python programs, specifically those of fifty lines or fewer, that have been transformed by common code modification tactics, and evaluate which tools are most appropriate for use in this environment. [ABSTRACT FROM AUTHOR]