Classification of single-trial self-paced finger tapping motion for BCI applications

Brain-Computer Interface provides a new communication paradigm between the human and machine, thus allowing physically impaired and paralyzed patients to control devices with the aid of brain activity alone, instead of using normal brain output pathways. In this paper, we present an algorithm to classify single-trial electroencephalogram (EEG) during the preparation of self-paced key tapping based on Common Spatial Subspace Decomposition (CSSD). Resulting 28 features for each trial from CSSD are classified using three classifiers 1) Linear Discriminant Analysis, 2) Quadratic Discriminant Analysis and 3) Support Vector Machine. For two class problem, linear subspaces are estimated using CSSD analysis that maximizes the variance of the signal for one class while minimizing the variance of the other. Improvement in the proposed work includes reduction in the number of features to 28 only that result in a significant decrease in computational complexity while improving the accuracy of classification from earlier reported 86% to 88% using Data Set IV of BCI Competition 2003 <self-paced 1s>.