An automated online proctoring system using attentive-net to assess student mischievous behavior

Area of Science:

• Computer Science
• Artificial Intelligence
• Educational Technology

Background:

• The shift to online and blended learning necessitates efficient remote examination monitoring.
• Traditional human proctoring methods for online exams are labor-intensive and resource-heavy.
• Scalability of online education is limited by the effectiveness of current remote evaluation methods.

Purpose of the Study:

• To develop and evaluate an automated, AI-based proctoring system for online examinations.
• To enhance the efficiency and reliability of remote proctoring through artificial intelligence.
• To address the limitations of human proctoring in large-scale online assessments.

Main Methods:

• Developed an 'Attentive system' utilizing AI for live video analysis of examinees.
• Integrated components for face detection, multiple person detection, face spoofing detection (Liveness net), and head pose estimation.
• Employed Attentive Net, Affine Transformation, face net, shallow CNN Liveness net, and SolvePnp equation for malpractice detection.

Main Results:

• The proposed automated proctoring system demonstrated high accuracy, reliability, and robustness.
• Achieved an improved accuracy of 0.87 through the combined use of Attentive Net, Liveness net, and head pose estimation.
• The system effectively detects various malpractices in real-time online examinations.

Conclusions:

• The AI-based 'Attentive system' offers a practical and scalable solution for automated online exam proctoring.
• This automated approach significantly reduces the need for manual labor and infrastructure.
• The system provides a reliable method for maintaining academic integrity in remote educational settings.