GAN-based novel feature selection approach with hybrid deep learning for heartbeat classification from ECG signal

Heart arrhythmias are one of the most important categories of cardiovascular illness. A heartbeat that is abnormal like too early, too slow, too fast, or uneven is indicated as an arrhythmia. Though some cardiac arrhythmias are benign, others can be dangerous and fatal if they are thought to be abnormal or the outcome of a damaged heart. The arrhythmias can be recognized by looking at and classifying the electrocardiogram (ECG) heartbeats. The automatic explanation of ECG data has witnessed a prominent development with the emergence of machine learning techniques. This paper develops an optimal deep learning technique to classify heartbeats. At first, pre-processing is done using median filter, resolution wavelet-based technique is exploited to recognize wave components. Subsequently, the features, like Discrete Wavelet Transform (DWT), autoregressive, Fractional Fourier-Transform (FrFT), and morphological features, are extracted. As the next step, feature fusion is performed by employing Kendall Tau, wrapper, and kraskov entropy together with Generative Adversarial Network (GAN). Lastly, heartbeat classification is done by employing proposed SExpHGS based DBN-VGG, where DBN-VGG is adopted by integration of Deep Belief Network and VGG, trained by employing Serial Exponential Hunger Games Search Algorithm (SExpHGS). Experimental outcomes illustrate that the SExpHGS based DBN-VGG approach performed superior when compared to conventional models with 95.7 % accuracy, 97.2 % sensitivity, and 94.9 % specificity rate.