In this Research, a new method has been proposed in order to classify the mental tasks which represent the Electroencephalogram (EEG) signal as time series. Time series are kind of data format which depict signal voltage varieties in time domain. Different parts of the different signals have different powers, so in first step and in the preprocessing, signal partitioning into several fixed windows is needed. Toward the extracting appropriate features from each EEG signal window, PCA algorithm is used. So for each window, a feature vector is made by PCA, and a general vector is created from these primary vectors. In order to refuse redundancy caused by non-important windows, the best combination of such vectors, that have the best results in classification, should be probed. Toward this goal, two feature extraction methods, Particle Swarm Optimization (PSO) and Genetic Algorithm (GA), are applied. K-Nearest Neighbor (KNN) was used as fitness function for PSO and GA. These methods select such windows whose combination of feature vectors are best and increase TP (true positive) of the classifier. The results show that GA and PSO improve the power of classification, but GA is more efficient.

Keywords