Modified LSTM-CNN Model for Arrhythmia Classification With Mixed Handcrafted Features

In this paper, we propose a novel machine learning approach for the classification problem of cardiac arrhythmias. First, we develop a combined model consisting of long short-term memory (LSTM) networks and convolutional neural networks (CNNs) to extract the deep features of ECG signals. Then, this LSTM-CNN model is augmented by mixed handcrafted features, including Heart Rate Variability (HRV) and morphological features. The final model is trained and validated over the MIT-BIH arrhythmia dataset. It can achieve a classification performance with an average accuracy of 99.58%, an average sensitivity of 99.42%, and average specificity of 99.62%. Experimental results demonstrate the effectiveness of the proposed approach.