TABLE 16.

Advantages and disadvantages of related works on ECG‐based cardiac arrhythmia classification using CNN.

Year	Author	Title	Algorithm	Dataset	Advantages	Disadvantages
2018	Özal Yıldırım, Paweł Pławiak, Ru‐San Tan and U. Rajendra Acharya	Arrhythmia detection using deep convolutional neural network with long‐duration ECG signals	Deep convolutional neural network	MIT‐BIH	1. Normalization. 2. Reduction of computational complexity. 3. Facilitate real‐time signal processing. 4. High accuracy of network training and validation	1. Feature extraction using end‐to‐end is not more useful for signals with high noise. 2. Not using data balancing in classes
2019	Nahian Ibn Hasan and Arnab Bhattacharjee	Deep learning approach to cardiovascular disease classification employing modified ECG signal from empirical mode decomposition	1. One‐dimensional deep convolutional neural network 2. Empirical mode decomposition (EMD) 3. Intrinsic mode functions (IMFs) 4. Stochastic gradient descent (SGD)	MIT‐BIH and PTB	1. Denoising. 2. Improving the accuracy of data validation. 3. Using learning rate scheduling to shorten training time. 4. Faster learning and reaching the highest accuracy with modified ECG	1. No adjusters (including core adjusters, bias adjusters, or activation adjusters) are used in any densely connected layer. 2. Overfitting. 3. The need for more training time to update the weights in each training step. 4. Failure to improve the proposed method with simultaneous training of IMF signals
2020	Dinesh Kumar Atal and Mukhtiar Singh	Arrhythmia classification with ECG signals based on the optimization‐enabled deep convolutional neural network	1. Optimization‐based deep convolutional neural network 2. BaROA‐based DCNN classifier algorithm	MIT‐BIH	1. More accuracy and efficiency in feature extraction. 2. Reducing the error in classification. 3. Higher convergence rate with a globally optimal solution. 3. The multi‐objective control capability of this system allows efficient arrhythmia classification	1. This method may require complex and expensive equipment when using sensors. 2. The use of the multiresolution wavelet‐based method can lead to higher computational complexity and increased demand for computing resources. 3. It has a higher tendency to avoid local optima mechanisms
2021	Marwa Fradi, Lazhar Khriji, Mohsen Machhout and Abdulnasir Hossen	Automatic heart disease class detection using convolutional neural network architecture‐based various optimizers‐networks	1. Convolutional neural network 2. Finite impulse response (FIR) filter	MIT‐BIH and PTB	1. Denoising. 2. Short processing time. 3. Using optimizers and improving cost performance. 4. Low computational complexity. 5. Increasing the learning rate and convergence speed	1. The implementation of a CNN architecture using a GPU can be limited by hardware and software resources. 2. Technical complications in implementing algorithms on GPU
2023	Sanjay Kumar, Abhishek Mallik, Akshi Kumar, Javier Del Ser and Guang Yang	Fuzz‐ClustNet: Coupled fuzzy clustering and deep neural networks for arrhythmia detection from ECG signals	1. Deep convolutional neural network 2. Fuzzy clustering algorithm	MIT‐BIH and PTB	1. Denoising. 2. Segmentation of the signals. 2. Balance the classes by augmentation. 3. Optimal feature extraction. 4. Increasing the accuracy of diagnosis	1. Absence of the 3/2 to 1/3 rule in training and testing. 2. Not using validation data to prevent overfitting in the training phase. 3. Computational complexity