This paper introduces a novel approach to biometric data security and performance optimization for face and iris recognition systems. Our methodology encompasses the extraction of features from face and iris data using state-of-the-art pretrained models, VGG16 and ResNet50. To safeguard the biometric data, we apply robust AES encryption, ensuring the protection of sensitive information against unauthorized access. The effectiveness of our approach is evaluated through a range of classification algorithms, including Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), and XGBoost. The integration of AES encryption with ResNet50 and SVM achieves an exceptional 99.80% accuracy, underscoring the method’s ability to provide both high security and superior performance. This work demonstrates the potential of combining advanced feature extraction, encryption, and classification techniques to enhance the reliability and security of biometric recognition systems.