Automated multi-class ECG arrhythmia detection using VMD and multi-task optimization

Electrocardiogram (ECG) classification is essential for accurately detecting and tracking heart rhythm disorders. This study proposes a multi-class ECG classification framework for identifying cardiac arrhythmias like Atrial Fibrillation (AF), Ventricular Fibrillation (VF), Normal Rhythm (NR), and Ventricular Tachycardia (VT). The ECG signals were decomposed using Variational Mode Decomposition (VMD), and higher-order statistics as well as entropy-based features were extracted from each mode. Multi-task Particle Swarm Optimization (MT-PSO) was employed to reduce redundant features and enhance the discriminative capability of the dataset. Multiple machine-learning models were evaluated, and optimized feature set led to clear performance improvements. The best results were obtained using LightGBM (ACC 0.993), HistGradientBoost (0.991), XGBoost (0.990), and ExtraTrees (0.990). Execution time also decreased for several models after optimization. Confusion-matrix and ROC analyses confirmed reliable detection across all four cardiac classes, and comparison with reported works shows that the proposed framework offers competitive or improved performance for ECG classification.