An ML-Assisted Intelligent Surveillance Proctoring and Cheating Detection System

Manual proctoring in large written exams often misses subtle cheating and overloads invigilators, while many AI proctoring tools raise concerns about privacy, bias, and opaque decisions. This study develops and evaluates an ML-assisted intelligent surveillance proctoring and cheating detection system for physical examination halls. The system uses Android devices and IoT cameras to capture student activity, applies on-device Google ML Kit detectors for head, hand, and object tracking, and sends selected frames to Google Gemini for higher-level behavioral analysis and explanation. The research follows an Agile software development life cycle and a quantitative Structured surveys using ISO/IEC 9126 design, including pilot deployment at the University of Science and Technology of Southern Philippines, structured ISO/IEC 9126 quality surveys with 49 teachers, and quantitative feedback on usability and trust. Results show an overall software quality rating of 3.9/5, with strong scores in reliability (4.2/5) and efficiency (4.3/5), and a 40–60% increase in detection of suspicious behaviors compared with manual proctoring alone. The system also preserves student privacy by keeping raw video on device and using incident-focused evidence logs, while keeping proctors in the decision loop. The core takeaway states: well-designed, policy-aligned AI support strengthens academic integrity in written exams while respecting student rights and supporting fair, documented decisions by human educators.