ECG Classification Using DTW - Based Learnable Kernels in Deep Neural Networks

Using Time Series Classification (TSC) methods in the study of biological signals like ECG for detecting unusual behavior is one of the most important applications of this field. With this motivation, we used kernel layer(s), as a novel approach, at the beginning of the common deep neural networks. These kernels have been trained based on Dynamic Time Warping (DTW) distance minimization. This new method has tested on two ECG datasets from UCR datasets: ECG200 and ECG5000 to classifying them. We got 91% and 92.3% accuracy for these datasets respectively, which is the best accuracy for ECG200 against other deep and non-deep methods and is an acceptable rate for ECG5000. Beside these results, the best achievement is the very low training time and also simplicity of the proposed network compared to other networks.