Feature Selection and Analysis EEG Signals with Sequential Forward Selection Algorithm and Different Classifiers

In this study, we investigated the features that could best represent EEG signals for brain computer interface systems and classifier accuracy was compared using different classification methods. EEG signals data set were taken from “BCI II Competition”. In this study, inadequate features that reduce classification accuracy were determined by using sequential forward selection algorithms and were extracted from real-dimensional feature matrix. The remaining active feature matrix and real-dimensional feature matrix were classified using k-nearest neighbor, subspace K-nearest neighbor, support vector machines, subspace discriminant and random forest decision tree algorithms. As a result of this study, the highest classification accuracy of real-dimensional feature matrix was obtained as 83.8% by random forest decision tree algorithm. In the other, the highest classification accuracy of dimention reductioned feature matrix with sequential forward selection algorthm was obtained as 96.4% by random forest decision tree algorithm.