Two comprehensive studies were conducted to evaluate this approach. In the first study, I applied biclustering to real-world assessment data consisting of dichotomously scored multiple-choice items and demonstrated the method’s superior performance over traditional approaches. Analyses further validated its robustness through simulations modeling a variety of realistic cheating scenarios. To enhance realism, the simulations also incorporated additional aberrant responses, such as rapid guessing due to time constraints and low motivation. In the second study, I extended the biclustering method to real-time cheating detection using mixed-format assessments, including dichotomous, polytomous, and multi-part items. This real-time application incorporated enhanced statistical significance testing to maintain low false positive rates and revealed strong detection performance across varying conditions and item types, including timestamp-based detection across different cheating scenarios.

Cheating detection in educational assessments remains a challenging issue, complicated by increasingly sophisticated cheating strategies and the rise of remote testing environments. In this dissertation, I introduce a novel biclustering approach for detecting cheating by simultaneously identifying groups of examinees and test items exhibiting suspicious response patterns. This biclustering method enables comprehensive detection of cheating behaviors by analyzing response accuracy, response time, and answer choices.

Results across both studies underscore the biclustering approach’s adaptability, computational efficiency, and precision, making it a valuable tool for maintaining test integrity in both post-exam analysis and real-time applications. I conclude by discussing practical implications, limitations, and directions for future research.