A study on the effect of psychophysiological signal features on classification methods

The most important factor affecting the performance of a BCI (Brain Computer Interface) systems, is classification feature set. Choosing the right features to increase the success of classification is the key point. In BCI systems, signals from brain are used to store into dataset. In this study, BCI Competition III dataset 1 consisting of ECoG (Electrocorticography) signals is preferred. In the first part, in order to decrease the processing load, the number of channels are reduced by eliminating channels (electrodes) which have low separation success. We developed new algorithm ADA (Arc Detection Algorithm) based on visual channel selection to determine quickly optimal channel subset. Than obtained Wavelet coefficients by Discrete Wavelet Transform (DWT) and determined classification features from Wavelet coefficients. These features are used to classify by KNN (K Nearest Neighbors), SVM (Support Vector Machine) and LDA (Linear Discriminant Analysis) by different feature set combination. The classification successes of feature combinations which are used in classification are compared. The impact on the classification performance of the right channel and the right property choice is observed. Test results are made with different frequency bands are compared with the same feature set. As a result, the highest classification accuracy of 95% was obtained by selected channels and feature.