An Overview of Machine Learning Approaches for ECG-Based Epileptic Seizure Detection

Electrocardiogram (ECG)-based epileptic seizure detection has emerged as a promising alternative to traditional EEG monitoring, particularly for long-term ambulatory applications. This paper presents a comprehensive analysis of ECG-based seizure detection methods spanning nearly five decades (1975-2024), with particular emphasis on machine learning approaches and their evolution. Through systematic analysis of 77 studies, we identify key trends in methodological development, evaluate performance across different algorithmic approaches, and provide insights into practical implementation considerations. Our analysis reveals that modern deep learning approaches achieve detection accuracies of up to 98.84%, while multimodal approaches combining ECG with other biosignals consistently out-perform single-modality methods. We present a structured framework for evaluating seizure detection algorithms and provide evidence-based recommendations for future development. Our findings indicate that careful feature selection and adaptive learning mechanisms are crucial for achieving robust performance in real-world applications.