<![CDATA[Manual attendance recording often leads to inefficiency and potential data manipulation. To address these issues, this study develops a web-based attendance system with real-time facial recognition using FaceNet. The model is designed by utilizing Dlib for face detection, FaceNet for facial feature embedding generation, and Support Vector Machine for identity classification. The system is implemented using Flask as the backend, Bootstrap for a responsive user interface, and SQLite as a lightweight database. The research dataset consists of 50 individuals and is used to compare the performance of three feature extraction models, namely FaceNet, MobileNet, and VGG-16. The evaluation results indicate that FaceNet achieves the best performance, with a training accuracy of 99.92%, a testing accuracy of 99.82%, an F1-score of 0.9983, and a training time of 9.74 seconds. MobileNet also demonstrates strong performance, achieving a training accuracy of 99.88%, a testing accuracy of 99.73%, and an F1-score of 0.9973. Meanwhile, VGG-16 shows relatively lower performance, with a training accuracy of 99.71%, a testing accuracy of 99.51%, an F1-score of 0.9951, and a training time of 24.61 seconds. These findings indicate that FaceNet is more effective and efficient in extracting facial features. Therefore, the developed system has the potential to replace conventional attendance methods that are prone to errors while supporting the implementation of a smart campus.]]>
Copyrights © 2026
