Evaluation of Feature Selection Methods for Improved EEG Classification

This paper compares several methods for feature selection used in EEG classification. Sequential, heuristics and population-based search methods are compared according to their efficiency and computational cost. A support vector machine classifier has been used to compare accuracies. Effect of the size of feature space has been explored by changing the total number of variables between 27 and 168. Experiments have been conducted to select channels as well as to select individual features from different channels.