A comparison of feature extraction methods for EEG signals

Feature extraction for automatic interpretation of EEGs has been extensively studied. A number of commercial approaches use exotic feature sets such as wavelets or nonlinear statistical measures such as fractal dimension. These choices of features were the results of evaluations and optimizations conducted on small research databases often collected under very controlled conditions. These approaches have not been extensively evaluated on big data or clinical applications using state of the art machine learning technology. Therefore, in this study, we compare performance of a number of standard feature extraction techniques on the publicly available TUH EEG Corpus using a state of the art classification system.