Mitigating Academic Dishonesty in Online Assessment using Machine Learning

The recent advancement in the field of information technology has resulted in the proliferation of online education all over the world. Much like traditional classroom education, assessments are an integral component of online education. During the online assessment, evaluation of the learning outcomes presents challenges mainly due to academic dishonesty among students. It results in unfair evaluations that raises questions about the credibility of online assessments. There exist several types of dishonesty in online assessments including exploiting the available Internet for finding solutions (Internet-as-a-Forbidden-Aid), illicit collaboration among students (Collusion) and third-party persons taking assessment on behalf of the genuine student (Impersonation). Several researchers have proposed solutions for addressing dishonesty in online assessments. These solutions include strategies for designing assessments that are resistant to cheating, implementing proctoring and formulating integrity policies. While these methods can be effective, their implementation is often resource-intensive and laborious, posing challenges. Other studies propose the use of Machine Learning (ML) for automated dishonesty detection. However, these approaches often lack clarity in selecting appropriate features and classifiers, impacting the quality of results. The lack of training data further leads to poorly tuned models. There is a need to develop robust ML models to detect different types of dishonesty in online assessments. In this thesis, we focus on Multiple Choice Questions (MCQ)-based assessments. We consider three types of dishonesty: (1) Internet-as-a-Forbidden-Aid, (2) Collusion, and (3) Impersonation prevalent in MCQ-based assessments. We developed individual ML models to detect students involved in each type of dishonesty during the assessment. The results also facilitate understanding the test-taking pattern of students and providing recommendations for cheat-proof assessment design. Finally, we present an Academic Dishonesty Mitigation Plan (ADMP) that addresses the diverse forms of academic dishonesty and provides integrity solutions for mitigating dishonesty in online assessments.