Pattern Recognition for Brain-Computer Interfaces by Combining Support Vector Machine with Adaptive Genetic Algorithm

Abstract

Aiming at the recognition problem of EEG signals in brain-computer interfaces (BCIs), we present a pattern recognition method. The method combines an adaptive genetic algorithm (GA) with the support vector machine (SVM). It integrates the following three key techniques: (1) the feature selection and model parameters of the SVM are optimized synchronously, which constitutes a hybrid optimization; (2) the aim of the hybrid optimization is to improve the classification performance of the SVM; and (3) the hybrid optimization is solved by using the adaptive GA. The method is used to classify three types of EEG signals produced during motor imaginations. It yields 72% classification accuracy, which is higher 8% than the one obtained with the individual optimization of the feature selection and SVM parameters.