Classification of L/R Hand Motor Imagery in Brain Computer Interfaces Using Feature Selection by Metaheuristic Algorithms

Pattern recognition field is necessary for the recognition of different sensorimotor tasks in Brain Computer Interface systems. Reducing the number of features is an important step in Brain Computer Interface systems and it can improve the accuracy and efficiency of the classification and reduce the costs.
In this paper, features selection was performed through using Improved Binary Gravitational search algorithm and Advanced Binary Ant Colony Optimization on data related to brain signals of nine normal subjects for imagination of left and right hand movements. Features were extracted from six different frequency bands. Two classifiers including support vector machine and k- nearest neighbor were applied to separate the classes. Data were processed by EEGLAB toolbox and through matlab software.
The classification rate of the proposed method is 84.21%. Using feature selection methods, effective frequency bands and features for left and right hand movement classification were extracted.
The results show the improvement in the classification rate by using Improved Binary Gravitational search algorithm and nearest neighbor classification.