Deep Learning based Arrhythmia Classification using Orthogonal Wavelet Filtering with Moth Flame Optimization Algorithm

An accurate and early detection of arrhythmia have become significant because of its enhanced mortality from the cardiovascular disease (CVD). The Electrocardiogram (ECG) majorly takes place in the CVD diagnosis, but it provides the information about the heartbeat. An automatic detection of arrhythmia performs a significant role in managing as well as curing the CVDs. In this research, pre-processing using an Orthogonal Wavelet-based Filtering approach with Moth Flame optimization (MFO) filter bank is proposed to enhance the denoising performance of the ECG signals. This research utilized the arrhythmia dataset named MIT-BIH to evaluated the performance of the proposed method. After pre-processing, the Convolutional Neural Network (CNN) is utilized for the feature extraction and Supervised Bidirectional Long Short-Term Memory (SBiLSTM) is used for the classification of the arrhythmia images. The proposed method attains better results and achieves the values of accuracy of 99.97%, specificity of 99.98%, sensitivity of 99.96% as well as Positive Predictive Value of 95.75% when compared to the existing methods like ResNet-BiLSTM-Attention, 2D-CNN, CNN-BLSTM, CNN-LSTM and Deep Recurrent CNN (DRCNN) respectively.