Multi-model fusion of Ornstein-Uhlenbeck process and RNN-XGBoost for biometric person identification using eye...

This paper introduces a novel framework for biometric person identification based on distinctive eye movement patterns. Grounded in foraging theory, the approach leverages the Ornstein-Uhlenbeck (O-U) process to model the dynamics of visual exploration and exploitation during gaze behavior. Eye movement data, including fixations and saccades, is analyzed using Bayesian estimation of a stochastic differential equation to extract individual-specific features. The extracted features are subsequently classified using a hybrid model combining Recurrent Neural Networks (RNN) and XGBoost. This multi-model fusion enhances robustness and discriminative capability. The method is evaluated using the publicly available FIFA eye-tracking dataset, achieving an average accuracy of 94% , F1-score of 94.03% , and an Area Under the ROC Curve (AUC) of 98.97%, with a corresponding Equal Error Rate (EER) of 4.0%.