Akademik Veri Yönetim Sistemi
Behaviormetrika, cilt.51, sa.1, ss.157-185, 2024 (Scopus)
In recent years, with the spread of large-scale online exams, the need for new methodological approaches to detect test cheating has increased. There are many psychometric-based approaches proposed in the literature for detecting test cheating. It appears that the absolute majority of currently proposed approaches focus on the derivation and development of statistics. Although these approaches can detect fraudulent test-taking behaviour with varying levels of detection rates, their practical use is somewhat limited, because these perspectives generally focus on a single class of methods, include only a particular type of variable as input, and address only one dimension of cheating behaviour. Alternatively, focusing on different classes of methods, multiple types of variables, and all dimensions of cheating behaviour can help provide highly accurate and versatile inferences. In this context, various holistic perspectives and supervised classification methods can offer functional solutions to this problem. Considering this point in the conducted research, it was aimed to investigate the effectiveness of the combined use of more than one method for detecting fraudulent behaviour in a test scenario, to determine the contribution that various supervised machine learning methods can make to exam security by evaluating them within a holistic process model with the assumption that some of the methods used in the literature are limited to one type of fraud. It is aimed to determine the effectiveness of taking these methods together in detecting test cheating in the best possible way by including many variables obtained throughout the test management into the system. Several holistic and combined methods that will be more effective and have higher detection power in detecting fraudulent test-taking behaviours have been examined with a real data set. It has been shown that the proposed approaches provide better fraud detection rates than other classical methods described in the literature.