EffNet: An efficient One-Dimensional Convolutional Neural Networks for efficient classification of long-term ECG fragments

Abstract Early Diagnosis of Cardiovascular disease (CVD) is essential to prevent a person from death in case of a cardiac arrhythmia. Automated ECG classification is required because manual classification by cardiologists is laborious, time-consuming, and prone to errors. Efficient ECG classification has been an active research problem over the past few decades. Earlier ECG classification techniques didn’t perform satisfactorily with greater accuracy and efficiency. An efficient 12-layer deep One-Dimensional Convolutional Neural Network (1D-CNN) titled EffNet is proposed in this research paper to automatically classify five distinct categories of heartbeats present in ECG signals. A unique collection of five different PhysioNet databases with ECG recordings of five different classes is created to enhance the dataset. These databases are segmented into ECG Fragments (long-term ECG signals of length 10-s) to effectively capture the ECG features between successive beats. These ECG fragments are then concatenated to form a merged dataset. Initially, sampling of the merged dataset is done. For balancing the dataset, Synthetic Minority Oversampling Technique (SMOTE) is used. Afterwards, 1D-CNN is employed with different sets of hyperparameters for the efficient classification of the ECG dataset. Classification of ECG of five different classes is also done through two deep Convolutional Neural Networks (CNNs), namely GoogLeNet and SqueezeNet, and Support Vector Machines (SVM). The statistical results obtained proved the dominance of EffNet over the transfer learning techniques (SqueezeNet and GoogLeNet) and SVM. Furthermore, a comparison is also made with the existing literature work carried out for ECG classification and the statistical results dominated over all others in terms of performance metrics.