Real-time Multi-Screen Cheating Detection using K-Means Clustering

Ensuring academic honesty during online exams is becoming more and more challenging with students taking advantage of multiple screens and mirrored monitors. This research presents a privacy-sensitive, real-time multi-screen behavior detection model that does not rely on cameras or biometric sensors. The system tracks hardware and behavioral signals like screen-switch rate, focus-loss activity, idle time, and display change events. Utilizing K-Means clustering (k = 3), these metrics are segregated into three categories: Normal, Suspected, and Cheating. Implemented in Python and tested on simulated and real datasets, the model registered a silhouette score of 0.27 and showed discriminative behavior segregation through clustering analysis. Testing against a labelled dataset produced balanced accuracy of more than 80 percent, supported by confusion matrix and performance curve research. Findings show that hardware monitoring and activity-based could be an effective, camera-free means of detecting cheating in online examinations. The approach is privacy-respecting, computationally light in real time, and has understandable output for administrative exam. Drawbacks include the focus to date on Windows platforms and the need for more comprehensive cross-platform testing. Future studies will examine multimodal integration and larger scales to further increase detection accuracy and transferability.