An Evolutionary Deep Learning Framework for Automated ECG Arrhythmia Classification

Abstract

This project presents a novel computational framework for cardiac arrhythmia classification that combines particle swarm optimization with convolution neural networks. The proposed system automatically optimizes neural network architectures for analyzing ECG signals to detect and classify multiple types of cardiac arrhythmias. The framework introduces a particle swarm optimization approach that autonomously determines optimal hyper parameters for the CNN architecture, eliminating the need for manual configuration. By leveraging the MIT-BIH Arrhythmia Dataset, the system demonstrates robust performance in classifying five distinct types of cardiac arrhythmias. The integration of evolutionary algorithms with deep learning enables automatic architecture optimization while maintaining high classification accuracy and minimizing categorical cross-entropy error. This innovative approach represents a significant advancement in automated ECG analysis by removing the dependency on manual hyper parameter selection, making it particularly valuable for clinical applications where expert knowledge of neural network design may be limited.