A Filtering Method for Classification of Motor-Imagery EEG Signals for Brain-Computer Interface

A brain-computer interface (BCI) utilizes brain signals such as electroencephalogram (EEG) and provides a path way for people to interact with external assistive devices. The objective of this work is to classify the tasks so that we can assist the disabled person in doing things on own way with the aid of BCI. The raw EEG signals have a chance of being affected with interference and hence have low signal to noise ratio (SNR) which may lead to erroneous results. These EEG signals are decomposed into intrinsic mode functions (IMFs) using different standard algorithms like empirical mode decomposition (EMD), multi variare empirical mode decomposition (MEMD). Different features like skewness, K-Nearest Neighbour (K-NN) entropy, sample entropy and permutation entropy are extracted from these IMFs which will significantly contribute to the classification of tasks. This work is carried out on the well established BCI motor imagery dataset, BCI competition IVa dataset-1 which will support the analysis. These extracted features are subjected to classifiers like random forest, Naive Bayes and J48 classifiers. The classification accuracies have been recorded and improved results are achieved using MEMD.