EEG-based motor imagery classification using wavelet coefficients and ensemble classifiers

Brain-computer interface (BCI) is a system that captures and decodes electroencephalogram (EEG) signals and transforms human thoughts into actions. To achieve this goal, using classification algorithms are most popular approach. However, classification of EEG signals can be categorized in complex problems because of high nonlinearity, high dimensionality, poor signal to noise ratio and poor spatial resolution. Combining classifiers is an approach to improve the performance of complex problems. In this article we studied the application of combining classifiers based on wavelet features to improve the performance of EEG signal classification in BCI systems. Three normal subjects K3b, K6b and L1b were asked to perform imaginary movements of left hand, right hand, tongue and foot during predefined time interval. EEG signals were decomposed into wavelet coefficients by discrete wavelet transform and used as feature vectors, presenting them into classifiers. Four combining classifiers were used to evaluate the EEG signals. Experimental results show that wavelet transform is an appropriate tool for the analyzing EEG signals. Also, according to the results of the experiments, mixture of experts overcomes the other used combining methods.