Motor imagery EEG signal classification on DWT and crosscorrelated signal features

Motor imagery (MI) based electroencephalogram (EEG) signals are a widely used form of input in brain computer interface systems (BCIs). Although there are a number of ways to classify data, a question still persists as to which technique should be employed in the domain of MI based EEG signals. In this paper, an attempt is made to find the best classification algorithm and feature extraction technique by comparing some of the prominently used algorithms on a same base dataset. Feature extraction techniques like discrete wavelet transform (DWT) and cross-correlation have been studied and compared. Five classification algorithms have been implemented which are logistic regression (LR), kernalised logistic regression (KLR), multilayer perceptron neural network (MLP), probabilistic neural network (PNN) and Least-square support vector machine (LS-SVM). Dataset IVa of BCI competition III has been used as a base dataset to test the algorithms. Evaluation of the algorithms has been done using a 10-fold cross-validation procedure. Experimental results show that a combination of DWT and LSSVM classifier outperforms the other procedures.