CNN-Based Cardiac Arrhythmia Classification from ECG

The electrocardiogram (ECG)-based arrhythmia classification plays a decisive role in diagnosing and clinical decision-making. In this study, the ECG arrhythmia classification using the MIT-BIH Arrhythmia Database has been proposed. The proposed robust method utilizes the features of discrete wavelet transform (DWT) denoising, class balancing, and multi-layer convolutional neural network (CNN) approaches. DWT denoising effectively reduces the noise interference in ECG signals and enhances the feature extraction quality. The multi-layer CNN architecture extracts and learns features from preprocessed ECG signals. The proposed network is trained on the balanced dataset to maximize classification performance. Experimental evaluation demonstrates that the average accuracy achieved is 99.97%. Moreover, the average sensitivity and precision reach 99.97% and 100%, respectively. The average loss of 0.0054 and overall F1-score of 99.98% signify the model's capability to minimize prediction errors.