Breaking up friendships in exams: A case study for minimizing student cheating in higher education using social network analysis

A well-known and persisting problem in modern education is academic dishonesty. There are various forms of such dishonesty, like plagiarism, which is often debated in the media, but cheating during examination perpetuates, and remains one of the oldest and most impactful forms of altering one's educational outcome and diminishing an institution's reputation. The applied prevention of this phenomenon is the subject of scientific attention, but the existing methods are most of the time insufficient or poorly applied. By analyzing the types of problems that occur during written exams, we have developed and implemented an innovative solution to decrease the amount of unwanted collaboration among students, by using their underlying friendship topology to the students' disadvantage. Consequently, we have introduced an original student placement strategy inspired by the interdisciplinary field of social networks analysis, and compared it to no placement strategy at all, and to the state-of-the-art random method. Our method is based on acquiring the social network of students participating in the exam, and using genetic algorithms to rearrange them in seats, such that there is minimal overlapping between real-world friendships and seated neighbours. The three methods have been applied independently on six different pools of students over the period 2013-2016, resulting in an extensive case study on N = 586 students in the Romanian higher education system. Next, we discuss the meaning of the results, as well as the applicability and limitations of our method. The analysis is presented both through empirical measurement of interaction between students during exam, as well as statistically, by introducing a metric for the placement effectiveness epsilon. Our proposed solution offers average improvements of x2.8 in terms of breaking up real-world friendships, and a x3.3 reduction in terms of empirically measured student interaction. On the other hand, we showcase that the easier to implement random placement brings about lower improvements of x1.7 (statistical) and x2.3 (empirically measured), over no seating strategy. Considering that many educational systems are unaware how vital the customization of student rearrangement is, we consider this case study to beacon an important institutional problem all around the world. (C) 2017 Elsevier Ltd. All rights reserved.