Siamese networks have emerged as a crucial tool in the field of machine learning, particularly for tasks that involve comparing pairs of inputs. This study investigates the implementation of Siamese networks using pairs of biometric data, such as fingerprints, keystroke dynamics, and photoplethysmograms, instead of conventional image pairs. The primary objective is to enhance biometric authentication systems by leveraging the unique characteristics inherent in biometric modalities. In this approach, we construct positive pairs (samples from the same individual) and negative pairs (samples from different individuals) to train the Siamese network. Each pair is processed through binary sub-networks that share weights, allowing the model to learn discriminative features that capture subtle differences in individual biometric traits. To further enhance our understanding of the model’s decision-making process, we will employ Explainable Artificial Intelligence (XAI) tools to elucidate the judgments made by the model.Our experiments demonstrate that this system significantly improves authentication sensitivity, reducing both false acceptance rates (FAR) and false rejection rates (FRR) compared to traditional methods. Additionally, we explore the impact of various biometric modalities on the performance of the Siamese network and investigate different architectures to optimize feature extraction. The results indicate that utilizing biometric pairs not only enhances verification accuracy but also increases the robustness of systems against spoofing attacks. This research contributes to advancing secure and effective biometric verification systems, making them more suitable for modern applications in security-sensitive environments while providing transparency through XAI tools to explain model decisions.