A feature extraction and classification algorithm for motor imagery EEG signals based on decision tree and CSP-SVM

In order to solve the problems of low recognition accuracy for motor imagery EEG signals, this paper presents a feature extraction and classification algorithm based on decision tree and CSP-SVM. Firstly, we select the fixed frequency of signals ranging from 8 to 30 Hz. Secondly, multiple spatial filters are constructed by using the one versus the rest common spatial pattern (OVR-CSP) and extract the feature vectors. Support vector machine (SVM) is employed to classify the feature vectors so that the best spatial filter is selected. We build the first branch of decision tree with the spatial filter selected and SVM. Then, OVR-CSP and SVM are used to build the branches of the decision tree repeatedly. Finally, 2005 BCI competition IIIa data set is used to validate the effect of the proposed algorithm. The results show that the highest accuracy of the proposed algorithm can reach 94.27% which proves the effectiveness of the algorithm.