ECG Signal as Robust and Reliable Biometric Marker: Datasets and Algorithms Comparison

Area of Science:

• Biometrics
• Signal Processing
• Machine Learning

Background:

• The need for secure and reliable user authentication is growing.
• Traditional biometric methods face challenges with spoofing and environmental variations.
• Electrocardiogram (ECG) signals offer a unique physiological signature.

Purpose of the Study:

• To investigate the efficacy of ECG signals as a biometric marker for authentication and identification.
• To assess the robustness of ECG-based biometrics against variations in data acquisition.
• To evaluate the performance of machine learning algorithms for ECG identification.

Main Methods:

• ECG signals were acquired from multiple sources with varying equipment and configurations.
• Several machine learning algorithms (LDA, KNN, MLP) were applied for identification.
• The impact of Principal Component Analysis (PCA) compression on identification accuracy was analyzed.

Main Results:

• ECG signals demonstrate high robustness to hardware variations, noise, and artifacts.
• The proposed method is stable over time and scalable to approximately 100 users.
• LDA, KNN, and MLP algorithms showed the most promising results for ECG identification.
• PCA compression did not improve, and sometimes reduced, identification accuracy.

Conclusions:

• ECG signals are a valid and reliable biometric marker for authentication.
• Machine learning algorithms, particularly LDA, KNN, and MLP, are effective for ECG-based identification.
• Data preprocessing techniques like PCA compression may not be beneficial for ECG identification.