A Simplified Tool for Testing of Feature Selection and Classification Algorithms in Motor Imagery of Right and Left Hands of EEG Signals

Some algorithms or a combination of them are more appropriated than others depending on the type of data that is being analyzed and what features and parameters are being used. In the analysis of motor imagery (MI) in an offline EEG-based brain-computer interface (BCI), different codes with different parameters are often used, making it harder to compare the effects of the algorithms applied. In this paper, we propose a simplified and limited tool that aims to aid in the testing of feature extraction, selection and classification algorithms separately or combined for the analysis of motor imagery in offline EEG-based brain signals while providing some information about the intermediate steps of a BCI construction. A known data set is used in order to ease the comparison between other researches. Only data from channels C3, Cz and C4 are used and the MI of left hand and right hand are analyzed. The data is filtered using a band-pass Chebyshev type II filter between 5 and 35Hz. Then, The rhythms mu and beta are isolated using a discrete wavelet transform (DWT) algorithm with a db4 mother wavelet of level 5. The proposed system has two outputs: the coefficients of the DWT related to the rhythms mu and beta; and a feature vector with three chosen features that can be used as an input to a classifier. The features extracted are mean, variance and energy. These are simple but effective features. Fixing some of the parameters simplifies the tool, offers a better environment for comparison of algorithms and allows the user to focus on specific steps of a BCI construction such as the feature selection and the classification phases.