Preventing unwanted access to data is the primary objective of information security. Passwords, user names, and keys are some of the most common techniques to verify a person's identity. All too easily stolen, misplaced, duplicated, or cracked, these traditional approaches have their limitations. A lot of focus is on multimodal biometric identification systems since they outperform their unimodal counterparts in terms of security and recognition efficiency. The low quality of biometric data is the main reason why single-modal biometric identification systems fail in actual public security operations. Low generalization and single-level fusion are two issues with present multimodal fusion approaches. In this paper, an innovative AI-powered multimodal biometric integration model DeepFuseNet is proposed that vastly improves accuracy and generalizability. Deep neural networks allow for the smooth integration of different fusion methods. Also, a virtual homogenous multimodal dataset has been built using synthetic operational information to verify the model's efficacy. In comparison to single-modal algorithms, multimodal feature fusion significantly improves experimental outcomes, leading to better enhancement in efficiency. © 2025 Elsevier B.V., All rights reserved.