ECG-HO-Net: A Hybrid Deep Learning Framework for Automated Arrhythmia Detection Using Hippopotamus Optimization and Liquid Neural Networks

With the increasing incidence of cardiovascular disorders and the limitations of existing ECG classification methods in handling signal variability and feature redundancy, there is a pressing need for a hybrid model that ensures both accuracy and adaptability in arrhythmia detection. The proposed ECG-HO-Net combines adaptive Liquid Neural Networks (LNN) and Hippopotamus Optimization (HO) for feature selection to detect arrhythmia. The process starts with preprocessing: Butterworth filtering removes noise, min-max normalization scales signals between 0 and 1, and Pan-Tompkins detects R-peaks to segment heartbeats. Feature extraction includes time-domain metrics such as Mean Heart Rate (99.82 BPM) and RMS (0.609), along with frequency-domain analysis using FFT to capture spectral components of the ECG. The model was evaluated on the MIT-BIH Arrhythmia dataset, comprising more than 109,000 annotated heartbeats of various types of arrhythmias. The novelty of ECG-HO-Net lies in its unique integration of HO for efficient feature selection and LNNs for adaptive temporal learning. Performance analysis shows that the proposed model outperformed state-of-the-art models, achieving 99.2% accuracy, 99.71% precision, 99.51% recall, 99.73% F1-score, and 99.24% specificity.