Time-frequency domain fusion and fractional adam optimized 1D CNN model for arrhythmia classification

Abstract Arrhythmias are characterized by irregular heart rhythms that pose a significant risk to human health. Therefore, timely and precise diagnosis is critical for effective treatment and management. This research proposes an enhanced approach for diagnosing arrhythmia using the Deep Learning (DL) technique and a new optimization method. First, the Discrete Wavelet Transform (DWT) is applied to filter the ECG information. Secondly, R-wave localization is performed by using the Daubechies wavelet. Third, the frequency information is extracted and fused with the Time-Frequency (TF) domain. The fused information is sent to the DL model, such as a Convolutional Neural Network (CNN). Specifically, we employed a lightweight one-dimensional CNN (1D CNN) architecture tailored to analyze electrocardiogram (ECG) signals. To increase the efficacy of 1D CNN, a new Fractional Adam Optimizer (FAO) is proposed to handle the complex and dynamic ECG data. The experiments have been conducted to evaluate the proposed model using the MIT-BIH and INCART12 arrhythmia datasets. The experimental outcomes show that the proposed method outperforms existing approaches by obtaining 99.71% accuracy. The experimental findings have been validated by using ten-fold cross-validation. The combination of TF domain fusion, 1D CNN, and FAO improves arrhythmia diagnosis accuracy and helps us understand heart dynamics. Overall, the proposed model leads to improved decision-making and better patient outcomes.