Exploring data augmentation methods to enhance EEG measures for epilepsy seizure detection

Automatic seizure detection using machine learning can reduce the workload of clinicians in epilepsy diagnosis. However, the class imbalance between seizure and non-seizure data limits model performance. Data augmentation offers a solution, yet few studies have systematically compared different augmentation strategies for seizure classification. In this study, we evaluate 12 data augmentation methods across multiple classifiers using EEG data. Beyond accuracy, we assess waveform preservation, spectral consistency, runtime, and feature separability. Results show that Magnitude Warping (MagWarp), Scaling, and Scaling for Multiple Channels (ScalingMulti) consistently yield superior performance. These findings provide practical insights into selecting effective augmentation techniques for real-world epilepsy detection systems and can support more reliable, automated diagnostic tools in clinical practice.