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

Area of Science:

• Neurology
• Machine Learning
• Biomedical Signal Processing

Background:

• Automatic seizure detection using machine learning aids epilepsy diagnosis and reduces clinical workload.
• Class imbalance between seizure and non-seizure data is a significant challenge impacting model performance.
• Systematic comparisons of data augmentation strategies for seizure classification are limited.

Purpose of the Study:

• To systematically evaluate the effectiveness of 12 data augmentation methods for seizure classification using electroencephalogram (EEG) data.
• To assess augmentation techniques based on accuracy, waveform preservation, spectral consistency, runtime, and feature separability.
• To provide practical insights for selecting optimal augmentation strategies for automated epilepsy detection systems.

Main Methods:

• Evaluation of 12 distinct data augmentation techniques applied to EEG datasets.
• Comparison of augmentation performance across multiple machine learning classifiers.
• Assessment of key metrics including accuracy, waveform fidelity, spectral characteristics, computational time, and feature discriminability.

Main Results:

• Magnitude Warping (MagWarp), Scaling, and Scaling for Multiple Channels (ScalingMulti) demonstrated consistently superior performance across evaluated metrics.
• These augmentation methods showed effectiveness in addressing class imbalance issues in EEG seizure data.
• The study identified specific augmentation techniques that balance performance gains with data integrity.

Conclusions:

• Magnitude Warping, Scaling, and Scaling for Multiple Channels are recommended as effective data augmentation strategies for EEG-based seizure detection.
• These findings support the development of more reliable and accurate automated diagnostic tools for epilepsy.
• The systematic evaluation provides a practical framework for researchers and clinicians implementing machine learning in epilepsy diagnosis.