AI-Based Smart Attendance and Behaviour Monitoring System

Abstract

Traditional attendance systems suffer from time inefficiency, susceptibility to proxy attendance, and a complete absence of engagement insights. This paper presents an AI-based smart attendance and behaviour monitoring system that unifies facial recognition with real-time behavioural analysis into a single pipeline. The proposed system employs Multi-task Cascaded Convolutional Networks (MTCNN) [1] for robust face detection, FaceNet [2] for generating 128-dimensional facial embeddings, and a custom Convolutional Neural Network trained on a hybrid dataset for classifying pedagogically relevant emotional states. Experimental evaluation demonstrates an attendance recognition accuracy of 96.8% in multi-person classroom scenarios and approximately 90% accuracy in engagement-related emotion detection. A Flask-based web dashboard [3] provides real-time monitoring and comprehensive analytical reporting. Deployment across live classroom environments confirms that the system recovers 5–10 minutes of instructional time per session, eliminates proxy attendance, and supports data-driven pedagogical interventions — demonstrating both technical reliability and institutional practicality for real-world academic deployment.