HYBRID GENERATIVE ADVERSARIAL NETWORK AND REINFORCEMENT LEARNING FRAMEWORK FOR ROBUST ECG ANOMALY DETECTION

Abstract Electrocardiogram (ECG) anomaly detection is vital for diagnosing cardiovascular conditions such as arrhythmias and myocardial infarctions. Traditional methods often struggle with challenges like data imbalance, gradient instability, and limited generalization. In this study, we propose a novel Adaptive Gradient-Free Whale Optimization (AGWO) framework that combines metaheuristic-inspired neural tuning, adaptive whale optimization, and deep ensemble learning. The AGWO framework enhances performance by integrating gradient-free parameter tuning, dynamic nature-inspired optimization, and robust ensemble learning of CNNs, LSTMs, and GANs. Experimental results on the MIT-BIH Arrhythmia and PTB Diagnostic ECG datasets demonstrate significant improvements over state of the art methods, achieving an accuracy of 95.2%, sensitivity of 93.8%, specificity of 96.4%, and AUROC of 0.97. This innovative approach addresses critical challenges in ECG anomaly detection, offering robust, generalizable, and clinically viable solutions for real-time cardiac monitoring.