Improved Deep Learning-based Multiple Biometric Authentication System for ATM Transaction solving Cost and Ambiguity Problems of Multiple Card Usage
N. Thillaiarasu, M. Ravi Prasad · International Journal of Information and Computer Security · 2025
When considering the ATM machine, the multimodal authentication is performed. Here, the authentication using biometric information is achieved by applying the multi-channel adaptive EfficientNetV2 with attention mechanism (MAEv2-AM) each channel carries information of each image from retina, fingerprint, face and spectrogram), where the hyper-parameters is tuned by using the improved fitness condition-based Archimedes optimisation algorithm (FCAOA). After the initial authentication, a single PIN entry would be required to select the desired bank account, streamlining the process of withdrawing money. This approach enhances convenience while maintaining security. Therefore, wastage of cards can be reduced, cost of card manufacturing can be reduced, and the ambiguity of using multiple passwords in multiple cards can also be reduced. Finally, the performance of the method is analysed using diverse metrics and compared with other baseline models. Hence, the proposed work outperforms with the higher results to do the biometric-based authentication for ATM transaction.