Classification of electrocardiogram signals with support vector machines and particle swarm optimization

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Electrocardiogram (ECG) analysis is crucial for diagnosing heart conditions.
• Automatic classification of ECG beats faces challenges like high dimensionality and limited training data.
• Support Vector Machine (SVM) shows promise but requires optimization for improved generalization.

Purpose of the Study:

• To experimentally validate the superior generalization capability of SVM for automatic ECG beat classification.
• To propose and evaluate a novel classification system, particle swarm optimization-SVM (PSO-SVM), for enhanced ECG beat classification performance.
• To assess the performance of SVM against k-nearest neighbor (kNN) and radial basis function (RBF) classifiers.

Main Methods:

• Conducted experiments using ECG data from the MIT-BIH arrhythmia database.
• Classified five types of abnormal waveforms and normal beats.
• Optimized SVM parameters and feature subsets using particle swarm optimization (PSO).
• Compared PSO-SVM with SVM, kNN, and RBF classifiers under varying training data sizes.

Main Results:

• The proposed PSO-SVM system achieved an average accuracy of 89.72% on 40,438 test beats.
• PSO-SVM significantly outperformed SVM (85.98%), kNN (83.70%), and RBF (82.34%) classifiers.
• The study confirmed SVM's superiority over traditional classifiers and highlighted PSO's effectiveness in improving SVM performance.

Conclusions:

• The SVM approach demonstrates superior generalization for ECG beat classification compared to kNN and RBF.
• The novel PSO-SVM classification system offers substantial improvements in accuracy.
• The findings suggest PSO-SVM as a highly effective method for automated ECG analysis.