Classification of ECG signals using multi-cumulants based evolutionary hybrid classifier.

Area of Science:

• Biometrics
• Signal Processing
• Machine Learning

Background:

• Electrocardiography (ECG) waveforms are unique to individuals, offering potential for human identification.
• Existing ECG-based identification methods require robust and accurate classification techniques.
• ECG databases often present challenges due to imbalance and noise.

Purpose of the Study:

• To develop a novel, robust, and accurate approach for human identification using ECG signals.
• To enhance the performance of ECG-based biometrics through advanced signal processing and machine learning.
• To introduce an "evolutionary hybrid classifier" for ECG data analysis.

Main Methods:

• Utilized MLII, UCI repository arrhythmia, and PTBDB ECG databases, addressing data imbalance with resampling techniques.
• Applied discrete wavelet transform (DWT) for noise removal and extracted features using multi-cumulants.
• Classified multi-cumulant features with kernel extreme learning machine (KELM), optimizing parameters via genetic algorithm (GA).

Main Results:

• Achieved 100% classification accuracy on MLII and UCI repository arrhythmia databases.
• Obtained 99.57% classification accuracy on the PTBDB database.
• Demonstrated superior performance compared to existing state-of-the-art approaches.

Conclusions:

• The proposed evolutionary hybrid classifier is highly effective for human identification using ECG signals.
• Multi-cumulant feature extraction combined with optimized KELM provides a robust biometric solution.
• The approach offers a significant advancement in ECG-based biometrics for accurate and fast identification.