Improvement of the classification achievements of ECG signals using extreme learning machines

Electrocardiogram (ECG) is a biomedical signal that provides information about the electrical activity of a heart. The term non-stationary is often used in signal processing to describe a process in which the spectrum of the signal changes over time. The ECG signal provides cardiologists with useful information about the rhythm and operation of the heart. Analysis of the ECG signal is an effective method used to detect and treat different types of heart diseases. The Cardiologists identify the typical waveforms of a patient's ECG signal and form the basis for diagnosis. The clinical applications of ECG are performed in a subjective manner, depending on the screening of the morphological features specified in the various guidelines by the naked eye. Computer-aided automated ECG analysis is seen as a promising method to ensure a more quantitative and consistent ECG signal interpretation. Therefore, it is important to accurately detect ECG signals and to classify them with a high success rate. In this thesis, the use and development of Extreme Learning Machine (ELM) algorithms were performed in the analysis of ECG signals which is the non-stationary signal. The analysis of ECG signals consists of three steps. These are Pre-processing, Feature Extraction and Classification process. In the preprocessing process, baseline wandering and 60 (Hz) noise of the ECG signal have been corrected by using the wavelet family symlet10 wave, combo, and Savitzky-Golay filters. MIT-BIH and PTB ECG datasets available in the PhysioNet database which is open access for feature extraction and classification process have been used. In the thesis work, the traditionally used morphological features have been taken into consideration primarily. At the same time, statistical features of ECG signals have also extracted. Morphological features of the ECG signal have obtained using Pan-Tompkins and Discrete wavelet transform. In this context, a total of 11 feature have obtained from ECG signals. In order to carry out the classification process within the scope of experimental studies, Artificial Neural Network, Extreme Learning Machine (ELM), Support Vector Machine (SVM), k-Nearest Neighbors (k-NN), and Genetic Algorithm based Wavelet Kernel Extreme Learning Machine Algorithm (GAWKELM) and Differential Evolution Algorithm based Extreme Learning Machine (DEA-ELM) have been used. The performance values of the classifiers have compared with each other. In this context, Wavelet kernel ELM based on Genetic algorithm method has been added to the literature. The genetic algorithm used in the wavelet kernel ELM has been shown to contribute significantly to achieving highly promising results by optimizing the α, β and θ coefficients in the wavelet kernel ELM combined with classical ELM. The most efficient results were obtained 3, 9 and 4 optimum wavelet values and accuracy 90% by using Differential development kernel algorithm and Grid search methods. In addition, the highest classification performance value has reached by using Differential Evolution Algorithm with 97.5 %.