<![CDATA[Student attendance remains susceptible to proxy attendance and operational inefficiencies, while many face recognition systems are evaluated only on benchmark datasets and rarely investigate identity representation strategies under real-world educational conditions. This study evaluates a hybrid MTCNN–ArcFace framework incorporating centroid-based identity representation, application-specific threshold calibration, and real-time validation for automated student attendance. A quantitative experimental design was conducted using a locally collected dataset from a secondary school. MTCNN was employed for face detection and alignment, whereas ArcFace with a ResNet-50 backbone generated facial embeddings that were aggregated into centroid templates for identity matching. The framework was assessed through offline performance evaluation and operational deployment. The proposed approach achieved 92.86% accuracy, 97.22% precision, 92.86% recall, and a 93.49% F1-score, with a 4.76% false acceptance rate and 2.38% false rejection rate. In addition, centroid representation reduced template storage requirements and supported efficient real-time recognition using limited enrollment samples. These findings demonstrate that centroid-based identity representation enhances the practicality of deep face recognition for educational attendance systems by improving computational efficiency while maintaining reliable recognition performance in authentic school environments.]]>
Copyrights © 2026
