Semi-Supervised Epileptic Seizure Detection Using Anomaly Detecting Autoencoder

Epilepsy is a neurological condition characterized by recurrent, unpredictable disruptions in brain activity, commonly manifesting as epileptic seizures (ES). A key method for seizure detection involves analyzing ElectroEncephaloGraphic (EEG) signals, which are inherently complex and highly variable. While Machine Learning (ML) and Deep Learning (DL) approaches have been increasingly applied for automated EEGbased seizure detection, ML models often demand extensive feature engineering, whereas DL models, despite their strong performance, typically suffer from high computational complexity and dependency on large datasets. In this work, we propose a lightweight channel-wise anomaly detection model based on a convolutional autoencoder architecture with only 110.59 k parameters, significantly reducing computational overhead. Our model uses reconstruction errors to effectively distinguish abnormal EEG patterns without the need for handcrafted features or massive datasets. The proposed approach achieved a detection accuracy of 96.61%, sensitivity of 94.96%, precision of 98.20%, and F1-score of 96.55%, outperforming state-of-the-art methods in both accuracy and model efficiency. These results highlight the model's potential for deployment in real-time applications, such as wearable devices for continuous seizure monitoring, offering a promising balance between performance and resource efficiency compared to conventional deep learning approaches.