Effect of features extraction on improving LSTM network quality in ECG signal classification

摘要

This article focuses on the extraction of features extracted from ECG measurement signals to improve the quality of LSTM network operation. Two features were distinguished from each individual sequence of ECG signals: instantaneous frequency (IF) and spectral entropy (SE). Both of these features are extracted from ECG signals using short-time Fourier transform. The applied approach enables the conversion of original measurement sequences into spectral images, from which IF and SE coefficients are then generated. As a result of the research, it was found that feature extraction significantly improves ECG signal classification both in terms of forecasting accuracy and in terms of network learning speed.