A Framework for Integrated Bank ATM Transactions Security Based on Multi-Modal Biometric Authentication Using Adaptive and Attentive Assisted Mobilenet-V3

Abstract

In recent times, the Automated Teller Machine (ATM) scam has increased widely in this society. The technology has emerged to steal money or hack the service during the ATM transaction. Many thieves are using the strategy of skimming and trapping to steal money from ATM devices. In order to secure the transaction in ATMs, several authentication frameworks with recognition have been implemented. This safety work could be dealt with through biometric identification such as face, retina, and fingerprint recognition of the user. Due to high demand for security and reliable authentication schemes, the multimodal biometric system has emerged. The multimodal biometric system needs more than one biometric trait of an individual for identification and security purposes. Therefore, an integrated bank ATM transactions security model using a deep learning-based multi-modal biometric authentication system is developed to provide higher security during ATM transactions. Initially, the ATM card with Personal Identification Number (PIN) is given as the input for making ATM transactions in integrated banks. If the PIN is verified, then the appropriate person is identified through the biometric information of the user. The biometric information of the users includes Fingerprint, Face recognition, Retina, and speech. The speech information is in the format of signal, and hence the de-noising is performed to eliminate the noises from the input signal. The de-noised signal is given to Short Time Fourier Transform (STFT) to perform a signal transformation. After applying STFT, the spectrogram of images is attained. Finally, both the fingerprint, face, Retina, and spectrogram images are combined and given as the input for the recognition stage. Here, the Adaptive and Attentive-based Mobilenet-v3 (AAMNet) network is used for the recognition of input images, where the parameters from the Mobilenet-v3 are optimized using the Enhanced Archerfish Hunting Optimizer (EAHO) to improve the recognition performance. After recognizing the biometric information of the users, the money transactions in ATMs are completed. Therefore, the security of the integrated banking ATM transaction is highly improved, and the illegal transaction is avoided. The experimental result of the developed security model in ATM transactions is validated with the traditional models to ensure the effectiveness of the developed system. Hence, the effective results of the proposed model attain nearly 93% for accuracy, sensitivity, specificity, and also nearly 6% for FPR and FNR, respectively. This result could be used for practical applications like banking and the financial sector, money transactions, funding management, and so forth.