Time Series Classification of Electroencephalography Data

Electroencephalography (EEG) is a non-invasive technique used to record the electrical activity of the brain using electrodes placed on the scalp. EEG data is commonly used for classification problems. However, many of the current classification techniques are dataset specific and cannot be applied to EEG data problems as a whole. We propose the use of multivariate time series classification (MTSC) algorithms as an alternative. Our experiments show comparable accuracy to results from standard approaches on EEG datasets on the UCR time series classification archive without needing to perform any dataset-specific feature selection. We also demonstrate MTSC on a new problem, classifying those with the medical condition Fibromyalgia Syndrome (FMS) against those without. We utilise a short-time Fast-Fourier transform method to extract each individual EEG frequency band, finding that the theta and alpha bands may contain discriminatory data between those with FMS compared to those without.