EEG Dataset Reduction and Classification Using Wave Atom Transform

Brain Computer Interface (BCI) systems perform intensive processing of the electroencephalogram (EEG) data in order to form control signals for external electronic devices or virtual objects. The main task of a BCI system is to correctly detect and classify mental states in the EEG data. The efficiency (accuracy and speed) of a BCI system depends upon the feature dimensionality of the EEG signal and the number of mental states required for control. Feature reduction can help improve system learning speed and, in some cases, classification accuracy. Here we consider Wave Atom Transform (WAT) of the EEG data as a feature reduction method. WAT takes input data and concentrates its energy in a few transform coefficients. WAT is used as a data preprocessing step for feature extraction. We use artificial neural networks (ANNs) for classification and perform research with varying number of neurons in a hidden layer and different network training functions (Levenberg-Marquardt, Conjugate Gradient Backpropagation, Bayesian Regularization). The novelty of the paper is the application of WAT in the EEG data processing. We conclude that the method can be successfully used for feature extraction and dataset feature reduction in the BCI domain.