Leveraging FaceNet and SVM Hyperparameter Optimization in Face Recognition Attendance Systems with Limited Data

Implementing a reliable Face Recognition Attendance System is crucial for organizations to streamline attendance tracking. Traditional methods, such as fingerprint and RFID systems, face significant limitations. Fingerprint systems can be hindered by factors such as moisture, injury, or environmental conditions, which affect their reliability. Similarly, RFID systems, while convenient for contactless entry, are susceptible to misuse, as cards may be used by individuals other than the registered users. This research overcomes these challenges by employing MTCNN for face detection, FaceNet for feature extraction, and Support Vector Machine (SVM) for classification, focusing on environments with limited data. The developed model achieved an impressive 100% accuracy on cross-validation training and independent test sets, showcasing its robustness even when operating under conditions of limited data. Hyperparameter tuning of the SVM led to the identification of optimal parameters, specifically C = 1 and gamma = 1, which resulted in the model consistently achieving a perfect 100% accuracy during cross-validation. Comparative analysis with ResNet50 and VGG16 highlighted the superiority of the FaceNet+SVM model, which demonstrated perfect classification results across all data. In contrast, ResNet50 and VGG16 exhibited signs of overfitting, failing to maintain accuracy on independent test data.