AI-based Surveillance for Exam Integrity: Real-Time Detection of Abnormal Student Behavior

The upholding of academic integrity during testing/assessment is vital and so is noticing abnormal student behaviors in examination halls in combating abuses. The assessment of this aspect tangibly most often concerns human invigilators which is conscious of human limitation and in a lot of cases subtend inefficiency, thus revealing a huge gap in research. This work intends to present a solution to this gap by creating a behavior detection system that is powered by artificial intelligence and is able to monitor in real time the activities of the examination hall and to raise alarms whenever suspicious activities are detected using the YOLOv8 model which is the latest of the object detection models. The procedure entails teaching the model with a specialized database that includes different testing environments, for example, the position of participants' heads and presence of restricted materials. The preliminary findings illustrate the remarkable capability of the model in recognizing and alarming invigilators of such activities as spies and cell-phone operators, while also allowing looking away from the screen towards other undesired directions. Assurance of fairness and credibility depends on academic integrity during tests. Conventional approaches of monitoring based on human invigilators are prone to mistakes and inefficiencies. This work suggests a YOLOv8-based artificial intelligence-based surveillance system to instantly identify aberrant student behaviour. The model is trained on a specialised dataset spanning several exam situations including head motions and illegal device use. Relevant for live monitoring, the system achieves 85.34% precision, 65.01% recall, and a mAP50 of 80.10%. Along with techniques for improving accuracy and dependability, we address issues including environmental variances, data bias, and real-time processing limits. Future research will concentrate on improving model accuracy and including other behavioral indicators.