<![CDATA[The type of arrhythmia can indicate the location of the disorder and its causes. The way to identify the arrhythmia is to use an electrocardiogram (ECG) strip. Machine learning can be used as an approach to assist identification of arrhythmias through an ECG. Extreme Learning Machine (ELM) is one single-hidden layer feedforward neural networks (SLFNs) that can be used for the classification of arrhythmias in order to assist medical diagnosis. To optimize ELM performance, Fast Independent Component Analysis (FastICA) algorithm is used for preprocessing and extracting ECG signals. In this study, several parameter tests were conducted to determine the impact on the performance of the classification model. ECG data obtained from the arrhythmia database managed by the Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH). Each data is a 3 seconds ECG snippet with total of 210 data divided into 6 arrhythmia classes and normal rhythms. The results showed that the classification model was able to achieve perfect performance with accuracy, precision-recall, and F-1 score of 100% at the training stage. However, the classification model was experiencing overfitting at the testing stage with the mean of matthew correlation coefficient is approximately 0. Overfitting occured because the feature representation is too complex and not proportional to the amount of available data. This resulted in poor performance in the ELM-FastICA test for data that was not yet recognized.]]>
Copyrights © 2019
