Using Biclustering to Detect Cheating in Real Time on Mixed-Format Tests.

Area of Science:

• Educational Measurement
• Data Mining
• Computational Statistics

Background:

• Traditional methods struggle with complex, mixed-format assessments.
• Real-time detection of academic dishonesty is crucial for maintaining assessment integrity.
• Biclustering offers a flexible approach to identifying unusual response patterns.

Purpose of the Study:

• To evaluate a real-time biclustering method for detecting cheating on mixed-format assessments.
• To assess the method's accuracy, computational efficiency, and adaptability across various testing conditions.
• To incorporate enhanced statistical tests for improved real-time detection and reduced false positives.

Main Methods:

• A real-time biclustering algorithm was developed and applied to simulated assessment data.
• The method jointly groups examinees and items based on response patterns.
• Enhanced statistical significance tests were integrated to refine detection accuracy.

Main Results:

• The biclustering method effectively detected cheating on both uniform and mixed-format assessments.
• Strong detection performance was observed across varying proportions of cheaters, group sizes, and compromised items.
• The method demonstrated high computational efficiency suitable for real-time applications with low false-positive rates.

Conclusions:

• Biclustering is an adaptable and versatile tool for real-time cheating detection in educational assessments.
• The enhanced method provides accurate and efficient identification of dishonest behavior across diverse item types.
• This approach supports maintaining academic integrity in digital and mixed-format testing environments.