Akademik Veri Yönetim Sistemi
10th International Conference on Electrical and Electronics Engineering (ELECO), Bursa, Türkiye, 30 Kasım - 02 Aralık 2017, ss.568-571
The purpose of Magnetoencephalography (MEG) based brain computer interface applications are to develop machine learning algorithms from small magnetic fields generated by neuronal activity and to make theoretical predictions from these algorithms. In this study, brain-computer interface competition 2008 data set 3 containing modulated MEG signals obtained by performing the imagination of right, left, forward and backward movements of the subjects' wrists were utilized. The aim of this study is to decide which two classes give better classification accuracy by using different classifiers. The signals recorded from this 4-class dataset were first reduced to 2 classes as forward-backward, right-forward, right-backward, left-forward, left-backward and right-left and then different feature extraction methods were applied to these combinations. The classification accuracy of the right-left direction gives best results with the random forest classification using the kurtosis of wavelet transform coefficients of the signals taken from channel 5 and 7 of the 10-channel data set. These two classes were investigated as the most discriminated two-class among the 4-class data set.