<p>Biometric security offers a robust approach to authentication and access control by verifying individuals based on unique physiological traits. Although Biometric Identification Systems (BIS) provide convenience and enhanced security, current single and multimodal biometric systems suffer from high error rates, limited adaptability, and susceptibility to spoofing attacks. These issues largely stem from the systems’ dependence on static authentication parameters and fixed keys. To overcome these limitations, this study proposes a novel biometric authentication framework named Separately Extracted Feature Fusion Convolutional Neural Network with Bat Optimization (SEFF-CNN-BO). The framework integrates three biometric traits—iris, face, and fingerprint—for enhanced security and reliability. These inputs undergo preprocessing to reduce noise and enhance quality, followed by feature extraction using Singular Value Decomposition (SVD) and Gabor filters. The extracted features are then fused into a unified vector through the SEFF-CNN-BO model. For key generation, the fused features are optimized using the Bat Algorithm and permuted based on user-provided input strings. The final cryptographic key is produced using SHA-256 hashing, ensuring strong security and revocability. To facilitate secure key sharing, Attribute-Based Encryption (ABE) is incorporated, enabling only authorized users to access the key. The proposed system was implemented in Python and evaluated using standard performance metrics, including precision, recall, accuracy, specificity, and F1-score. The results demonstrate superior performance, achieving 99.84% accuracy, 99.64% precision, 99.54% specificity, and 99.64% recall. Comparative analysis with conventional models confirms the enhanced effectiveness of the proposed approach. Furthermore, cryptographic key strength and revocability were validated using the NIST statistical test suite, affirming the system’s robustness and complete revocability.</p>