Cross-Condition Analysis of an SSVEP-Based Biometric Authentication System

As biometric authentication becomes increasingly important, EEG-based methods offer a promising alternative due to their resistance to spoofing and subject-specific characteristics. Among EEG paradigms, steady-state visual evoked potentials (SSVEPs) provide stable and high signal-to-noise responses with minimal user training. However, the performance of SSVEP-based biometric systems depends heavily on experimental design and model configuration, which remain insufficiently explored. In this pilot study, we systematically evaluate person authentication performance using three convolutional neural network-based models on two public datasets. We examine how training block count, trial duration, and temporal segmentation affect classification accuracy. Results show that these factors interact in complex ways, and model performance varies depending on the input design. Our findings provide insights into how design choices affect model performance, offering guidance for developing robust SSVEP-based biometric person authentication systems.