Ensemble Classifiers for Brain Cyborgs

Abstract

Brain Cyborgs is a new era of neuroscience to control machines by thoughts. It involves identifying brain regions for recording of signals to harvest the underlying control information to direct machines. Such technological advancement would bring hope to tetraplegic and quadriplegic patients living a locked-in life. But, its successful implementation has many technological gaps. The major challenge is identification of brain information and its decoding. Differentiation and classification of brain signals and their mapping to the corresponding mental state requires computational algorithms which are efficient, reliable and fast. Majority of machine learning algorithms revolve around single learners and classifiers. Another popular approach is to combine multiple classifiers to achieve final decision. In this approach decisions of individual classifiers are scaled to contribute to the final prediction. In this paper, classifiers have been trained using ensemble learning approach and evaluated on motor imagery (MI) EEG data. Three methods of ensemble classification viz. Boosting, Bagging and Random Subspace have been validated using time-frequency features. It has been found that ensemble classifiers are effective in classification of MI EEG signals with an accuracy of 85.83%.