Authors:
Julia Opgen-Rhein
1
;
Bastian Küppers
2
and
Ulrik Schroeder
3
Affiliations:
1
IT Center, RWTH Aachen University, Seffenter Weg 23, Aachen and Germany
;
2
IT Center, RWTH Aachen University, Seffenter Weg 23, Aachen, Germany, Learning Technologies Research Group, RWTH Aachen University, Ahornstraße 55, 52074 Aachen and Germany
;
3
Learning Technologies Research Group, RWTH Aachen University, Ahornstraße 55, 52074 Aachen and Germany
Keyword(s):
E-assessment, Cheating, De-anonymization, Stylometrics.
Related
Ontology
Subjects/Areas/Topics:
Computer-Supported Education
;
Information Technologies Supporting Learning
;
Learning Analytics
;
Learning/Teaching Methodologies and Assessment
Abstract:
Electronic exams are more and more adapted in institutions of higher education, but the problem of how to prevent cheating in those examinations is not yet solved. Electronic exams in theory allow for fraud beyond plagiarism and therefore require a possibility to detect impersonation and prohibited communication between the students on the spot and a-posteriori. This paper is an extension of our previous work on an application for detecting fraud attempts in electronic exams, in which we came to the conclusion that it is possible to extract features from source code submitted for tutorials and homework in a programming course. These can be used to train Random Forest, Linear Support Vector Machine, and Neural Network classifiers and assign the exams from the same course to their authors. The Proof of Concept was further developed and this paper outlines the experimentally determined requirements for the selection of training and test data and its pre-processing to achieve applicable
results. We achieve an accuracy larger than 89% on a set of source code files from twelve students and found that material from all parts of a programming course is suitable for this approach as long as it provides enough instances for training and is free of code templates.
(More)