A Changepoint Rule for Determining Bicluster Retention Cutoffs in Cheating Detection

Area of Science:

• Educational Measurement
• Data Mining
• Statistical Analysis

Background:

• Biclustering identifies groups of examinees with unusual response patterns for test cheating detection.
• A key challenge is determining the optimal number of biclusters to retain and examinees to flag.
• Retaining too many biclusters risks false positives; too few may miss genuine cheating.

Purpose of the Study:

• To propose and evaluate a changepoint-based retention rule for biclusters in test cheating detection.
• To provide a data-driven method for setting bicluster retention cutoffs.
• To compare the proposed rule against established benchmarks like the F1-score.

Main Methods:

• A changepoint detection algorithm was applied to the ordered sequence of bicluster p-values.
• The method was validated using two operational test forms with known cheating labels.
• A simulation study varied cheating type, test length, and item compromise proportion.

Main Results:

• The changepoint-based rule closely matched F1-score benchmarks in empirical analyses, showing similar sensitivity and slightly lower specificity.
• In simulations, the rule approximated F1 benchmarks but favored more conservative cutoffs.
• The method demonstrated potential for flagging relevant biclusters for review in cheating detection.

Conclusions:

• The proposed changepoint-based retention rule offers a data-driven approach to bicluster selection in test cheating detection.
• This method can aid in optimizing the balance between identifying cheating and minimizing false positives.
• The rule provides a practical tool for operational settings requiring efficient review of potential cheating patterns.