Behavioral Biometrics-Based Continuous Authentication in FinTech Apps Using Triplet-Loss Siamese Models
Abstract
Behavioral biometrics offers a promising avenue for continuous user authentication in FinTech applications, enhancing security without disrupting user experience. This study focuses on implementing a Triplet-Loss Siamese Neural Network (TLSNN) to distinguish legitimate users from impostors based on their unique interaction patterns. Existing authentication methods often rely on static credentials or periodic checks, which are vulnerable to session hijacking, device theft, and mimicry attacks. Moreover, many lack adaptability to evolving user behavior. The proposed framework uses a Triplet-Loss Siamese architecture that learns discriminative behavioral embeddings by comparing anchor, positive, and negative samples, effectively modeling subtle variations in user behavior. By continuously monitoring inputs like swipe dynamics, typing cadence, and gesture pressure, the system ensures seamless, real-time verification. This approach enhances the personalization and robustness of user profiles, allowing for early detection of unauthorized access without compromising usability.