CB-HDM: ECG signal based heart disease classification using convolutional block attention assisted hybrid deep Maxout network

Abstract

Cardiovascular diseases (CVD) stand as the leading cause of death globally. Diagnosis of CVD relies on electrocardiograms (ECGs), which record the heart’s activity. Adequate labeled data is essential for effectively training machine learning techniques, which may also demand considerable time and memory resources. To address these existing issues, this research introduces an improved detection technique for predicting CVD. Initially, data is sourced from the MIT-BIH arrhythmia database, followed by pre-processing using the Modified FIR filter model (M-FIR) to filter ECG signals. It is used to enhance signal quality and eliminate unwanted noise. M-FIR is used to preserve the specific features of the signal. Feature extraction utilizes the Radial Hilbert Function Transform Network (RHFTN) technique. It can extract the temporal and spectral signal features for effective CVD detection. The Improved Bald Eagle Search Algorithm (IBES) is employed to select features and minimize irrelevant features. Detection and classification of heart disease from ECG signals are achieved using the Convolutional Block Attention Assisted Hybrid Deep Maxout Network model (CB-HDM). Losses in the network model are mitigated by the Gazelle Optimization Algorithm (GOA). The evaluation results are simulated using Python. The proposed model achieved 98.81% accuracy, 96.18% precision, 96.87% recall, and a 94.39% F1 score. The proposed model demonstrates superior efficiency compared to other classifiers.