Abstract:
This research presents the design and implementation of a secure, AI-powered attendance system utilizing dual
authentication through facial recognition and QR code verification. The study addresses the prevalent issues of proxy
attendance and inefficiency in traditional manual methods by developing a robust, automated solution. The core of the
system is a deep learning model built upon the MobileNetV2, EfficientNetB0 architecture, employing transfer learning
and strategic fine-tuning. The research methodology evolved through several iterative model updates. Initial attempts
with a simple CNN and an imbalanced dataset of 25 classes yielded limited accuracy. Subsequent improvements involved
curating a balanced dataset of 30 individuals, integrating comprehensive data augmentation, and optimizing the training
pipeline with advanced callbacks like ReduceLROnPlateau and EarlyStopping. The final, optimized model achieved a
validation accuracy of 88.44%, demonstrating a significant enhancement over baseline performances. Each student is
uniquely associated with both facial data and a personally encoded QR code containing their identity. The system’s
dual-authentication mechanism requires a successful match between a live facial scan and the corresponding QR code
before logging attendance, thereby establishing a strong defense against fraudulent entries. This work concludes that
the integration of a fine-tuned MobileNetV2 model with a dual-factor authentication framework provides a scalable,
accurate, and fraud-resistant alternative to conventional attendance systems. The solution is particularly suited for
academic environments, offering a practical balance of security, automation, and operational efficiency.
