Classification and feature extraction of biological signals using Machine Learning Techniques

Recently, the interest in electrocardiogram (ECG) signal analysis has grown, as it has been seen as a saddle point in diagnosing cardiovascular disease. The ECG is a standard noninvasive method for diagnostic and routine monitoring of the heart. Neural networks were used for automatic disease identification. In this context, the main subject of this article is the classification of ECG signals for the identification of heart functioning problems. Secondarily, we analyze how different acquisition frequencies of the ECG signals lead to variation in neural networks performance. To this end, two data sets containing ECG signals were used: PTB and PTB-XL. Four neural networks architectures were compared in terms of performance: the first and the second are based on convolutional neural networks and the third and fourth are derived from the first two, by adding a new branch containing nonlinear features extracted from the ECG signals. On the PTB database, the best results were obtained with a convolutional neural network with feature injection, with an accuracy of 89.012% for 100 Hz acquired signals. The best results for PTB- XL were obtained with the same network with an accuracy of 85.111% and 100 Hz.