Interpretable evaluation of physiological signals for biometric identification.

Area of Science:

• Biometrics
• Signal Processing
• Machine Learning

Background:

• Biometric systems using physiological signals achieve high identification accuracy (IA) and low Equal Error Rate (EER).
• Existing research often prioritizes performance metrics over model characteristics.
• Understanding how signal features, population size, and sample availability impact biometric system reliability is crucial.

Purpose of the Study:

• To develop a unified framework for analyzing ECG, EEG, and PPG signals in biometric systems.
• To investigate the influence of signal features, population size, and sample availability on biometric performance and reliability.
• To enhance the interpretability and inform the development of biometric systems for real-world deployment.

Main Methods:

• Proposed a unified framework to analyze ECG, EEG, and PPG signals.
• Employed machine learning models for performance assessment and interpretability.
• Utilized SHapley Additive exPlanations (SHAP) to quantify feature importance.
• Conducted sensitivity analyses on population size and sample availability.

Main Results:

• Achieved identification accuracy (IA) consistently exceeding 96% across most datasets.
• Demonstrated lower Equal Error Rate (EER) compared to existing deep learning-based studies.
• SHAP analysis identified wavelet coefficients and PPG systolic features as highly discriminative.
• Sensitivity analyses provided insights into performance variations with population size and sample availability.

Conclusions:

• The proposed framework enables a deeper understanding of biometric system behavior beyond traditional metrics.
• Identified key discriminative features and influencing factors for robust biometric system design.
• Supports the development of secure, privacy-aware biometric systems for diverse deployment scenarios, including resource-constrained edge environments.