Feature selection algorithm for evoked EEG signal due to RGB colors

In this paper, a single trial classification is introduced for the Electroencephalography (EEG) signals evoked by RGB colors. The effectiveness of a single trial classification is an important step towards online classification of EEG signals. Signals are analyzed by Empirical Mode Decomposition (EMD) technique, and the last decomposition is used in the feature extraction stage. We investigate different feature extraction methods in order to find out the best method which can be used with colors dataset. These methods are: Event-Related Spectral Perturbations (ERSP), Target mean, AutoRegressive and EMD residual. In addition, we propose a new feature selection algorithm, which focuses on selecting the best features by studying the behavior of EEG components that appear due to the introduced color. We introduced a comparison between the classification results of using all extracted features, the results of using the selected features by the proposed algorithm and the results of using the selected features by recursive feature elimination algorithm, which is used by similar study. The proposed algorithm is proved with all the investigated feature extraction methods as the classification accuracies are increased. Support Vector Machine (SVM) is used in the classification process. We found that the execution time of using color's stimulus is only 0.23s, which is much less than the time which was required by any other stimulus such as imagery and spelling word presented in the previous researches. The best feature extraction method that gives the highest classification accuracy and can be used with real time BCI systems are Target Mean and EMD residual, as their accuracies are high and the computation time is very low.