Performance analysis of dimension reduction algorithms on EEG signals

Brain computer interface is a structure that allows systems to be controlled with signals from the brain. In this study, we investigated the features that could best represent the computer interface systems, different dimension reduction methods were applied to the feature matrices and the best classification method was chosen. EEG signals were taken from the data set III of the preparation of the “BCI III Competition” competition. Linear Discriminant Analysis, Stochastic Neighbor Embedding and Maximally Collapsing Metric Learning algorithms were applied to feature matrices as dimension reduction method. Extracted features are classified by k-Nearest Neighbor and Support Vector Machine methods. As a result, the size was reduced by the Linear Discriminant Analysis algorithm and the highest success rate was obtained as 90.2% from the EEG data classified by k-Nearest Neighbor algorithm.