Utilization of a hierarchical electrocardiogram classification model for enhanced biometric identification

Area of Science:

• Biometrics
• Artificial Intelligence
• Signal Processing

Background:

• Electrocardiogram (ECG) signals offer inherent security due to forgery resistance.
• ECG signal variability under physical and cognitive stress challenges consistent user identification.
• Existing methods struggle with dynamic ECG signal characteristics.

Purpose of the Study:

• To develop a robust 2-stage user identification system using ECG signals.
• To improve the accuracy of ECG-based identification under varying user stress states.
• To enhance the real-life usability of ECG biometrics.

Main Methods:

• A 2-stage system integrating ECG signals and user status information.
• Classification of ECG signal status in the first stage.
• Utilizing feature values in a second model for enhanced dynamic feature learning.
• Performance evaluation on CSU-BIODB and MIT-BIH ST Change databases.

Main Results:

• Achieved identification accuracies of 92.08% (CSU-BIODB) and 95.83% (MIT-BIH).
• Obtained f1-scores of 0.9207 (CSU-BIODB) and 0.9369 (MIT-BIH).
• Demonstrated significant accuracy improvements (9.3% and 36.76%) over existing single-stage models.

Conclusions:

• The proposed 2-stage model effectively addresses ECG signal variability under stress.
• The system significantly enhances the accuracy and practicality of ECG-based user identification.
• This research provides a strong foundation for deep learning in ECG signal processing for biometrics.