Cardiacnet: Cardiac Arrhythmia Detection and Classification Using Unsupervised Learning Based Optimal Feature Selection with Custom CNN Model

Abstract

The irregularity in the heartbeats caused cardiac arrhythmia, which resulted in serious health problems. This cardiac arrhythmia is monitored by electrocardiogram (ECG) signals. As a result, an accurate and timely analysis of ECG data can prevent serious health problems. However, the conventional manual prediction systems and artificial intelligence (AI) methods failed to detect the cardiac arrhythmia because they failed to extract the deep salient features from the ECG dataset. So, this research work implements a model named as CardiacNet, which is used to identify and classify cardiac arrhythmias from a MIT-BIH-based dataset. Initially, the pre-processing operation is performed to remove the non-linearities from the dataset. Then, unsupervised machine learning algorithm-based principal component analysis (UML-PCA) is used to extract the features of the pre-processed dataset. Further, the optimal feature selection operation is carried out using improved Harris Hawk’s optimization (IHHO), which is a naturally inspired model. Moreover, a customised convolutional neural network (CCNN) model performs the classification of various cardiac arrhythmia diseases using IHHO features. The simulation results show that the proposed CardiacNet resulted in accuracy of 97.57%, sensitivity of 98.29%, specificity of 97.97%, F-measure of 97.40%, precision of 98.66%, Matthew’s correlation coefficient (MCC) of 98.17%, dice of 98.96%, and Jaccard of 97.12%. The performance comparisons show that the proposed CardiacNet resulted in improved metrics over all existing methods.