Wavelet De-Noising and Genetic Algorithm-Based Least Squares Twin SVM for Classification of Arrhythmias

The automatic detection of cardiac arrhythmias is a challenging task since the small variations in electrocardiogram (ECG) signals cannot be distinguished by the human eye. We propose a fast recognition method to diagnose heart diseases that is less time consuming and achieves better performance by combining wavelet de-noising with a genetic algorithm (GA)-based least squares twin support vector machine (LSTSVM). First, adaptive wavelet de-noising is employed for noise reduction. Second, power spectral density in combination with timing interval features is extracted to evaluate the classifier. Finally, a GA, particle swarm optimization (PSO), and chaotic PSO are compared for parameter optimization of the proposed directed acyclic graph LSTSVM multiclass classifiers. ECG heartbeats taken from the MIT-BIH arrhythmia database are used to examine the proposed method and other traditional classifiers such as multilayer perception, probabilistic neural network, learning vector quantization, extreme learning machine, SVM, and current TWSVMs. Number of our training samples is < 3.2 % of all samples. Our proposed method demonstrates a high classification accuracy of 99.1403 % with low ratio of training and testing sample sizes; furthermore, it achieves a more rapid training and testing time of 0.2044 and 55.7383 s, respectively.