Multimodal biometric authentication for social e-governance using blockchain with a differential privacy-based deep learning model

Today, biometric authentication has become a central component of user security in social governance systems, where each government department demands access to user-specific data that varies across agencies. However, storing such data in centralized repositories increases serious privacy concerns, as unrestricted access by multiple entities maximizes the risk of data leakage. To address this, our research presents a novel biometric authentication system integrating robust privacy-preserving techniques, built on advanced deep learning architectures and differential privacy algorithms. A blockchain ledger integrated with a Merkle tree is used to securely store user identities, providing tamper-evident cryptographic validation of registered users. We further develop a novel hybrid model by integrating a pre-trained Vision Transformer (ViT) with a differential privacy-based machine learning enhanced training strategy, wherein the model is trained on noise-induced images to resist inference attacks. The system without differential privacy achieves 90.80% accuracy, 0.94 precision, 0.91 recall, and an F1-score of 0.90 in the standard configuration, while the differentially private model maintains 68.97% accuracy with ε = 6.2, ensuring a strong privacy—accuracy balance. The evaluation confirms that our proposed model, incorporating differential privacy, provides a secure and scalable solution for managing sensitive citizen data while achieving reliable performance in privacy-aware biometric verification for real-world e-governance applications.