Real coded GA-based SVM for motor imagery classification in a Brain-Computer Interface

The brain signals are generally measured by Electroencephalogram (EEG) in Brain-Computer Interface (BCI) applications. In motor imagery-based BCI, the performed MI tasks (e.g., imagined hand movement) are identified through a classification algorithm to communicate and control the device. Consequently, improving the performance of the classifier is crucial to the success of the BCI system. One of the most popular linear classifier in BCI applications is the Support Vector Machine (SVM). This paper improves the performance of MI-based BCI by finding the optimum free kernel parameters of the SVM classifier. A real-coded genetic algorithm is utilized to determine the free kernel parameters of the SVM. The performance of this method is evaluated using publicly available BCI Competition IV dataset IIa for right and left hand motor imagery tasks. The results show that using real-valued GA-based SVM with Polynomial or Gaussian kernel improves the average accuracy over nine subjects compared with the baseline (i.e., the grid search method). Hence, using automated method (GA) helps us in improving the performance of the MI-based BCI especially for subjects with poor performance.