Deep Learning-Based Liveness Detection to Prevent Biometric Payment Fraud

Abstract

Biometric authentication has become a central pillar of secure digital payments, offering an efficient and user-friendly replacement for passwords and PINs. However, the widespread use of facial recognition and fingerprint scanning has also introduced a new attack vector—presentation attacks, including spoofing with masks, photos, and deepfake-generated artefacts. This paper explores a robust deep learning-based liveness detection framework designed to ensure that biometric verification systems can reliably distinguish between live users and spoofed attempts in real-time payment environments. The proposed model integrates convolutional neural networks (CNNs), attention-weighted feature extraction, and multi-modal temporal fusion to improve generalization and prevent high-risk fraudulent transactions. Experimental evaluation on widely used liveness-testing datasets, complemented by real-world payment terminal simulations, demonstrates strong performance, achieving high accuracy, reduced false acceptance rates, and stable inference under environmental variations such as lighting and camera quality. The findings emphasize the urgent need for improved biometric security in modern financial workflows and position deep learning-based liveness detection as a scalable, high-precision security layer for safeguarding biometric payment systems.