Support Vector Channel Selection in BCI

Abstract

Abstract. Designing a Brain Computer Interface (BCI) system one can choose from a variety of features that may be useful for classifying brain activity during a mental task. For the special case of classifying EEG signals we propose the usage of the state of the art feature selection algorithms Recursive Feature Elimination [1] and Zero-Norm Optimization [2] which are based on the training of Support Vector Machines (SVM) [3]. These algorithms can provide more accurate solutions than standard filter methods for feature selection [4]. We adapt the methods for the purpose of selecting EEG channels. For a motor imagery paradigm we show that the number of used channels can be reduced significantly without increasing the classification error. The resulting best channels agree well with the expected underlying cortical activity patterns during the mental tasks. Furthermore we show how time dependent task specific information can be visualized.

Toward enhanced P300 speller performance

“…It should be noted that SWLDA actually performs pruning since feature from all input channels are not guaranteed to be weighted by this procedure. Alternative feature extraction methods in a BCI context are given in [12] [14]. Additional evidence for the efficacy of using electrodes that include posterior, central, and frontal electrodes is provided in [21].…”

Classifying Event-Related Desynchronization in EEG, ECoG and MEG Signals

“…The sensors ranked most consistently highly are to be found in lateralized central and pre-central regions, bilaterally for the EEG experiment and for subject 308, and with a right-hemisphere bias for the others. For further examination of the performance of Recursive Channel Elimination in the identification of relevant source locations, see Lal et al (2004), Lal et al (2005c) and Lal et al (2005a).…”

“…Three experiments form the basis for this chapter: one using EEG (described in more detail by Lal et al, 2004), one using ECoG (Lal et al, 2005c), and one based on MEG recordings (Lal et al, 2005a).…”

“…For a selection of EEG motor imagery studies, see Wolpaw et al (1997); Birch et al (2003);McFarland et al (1997); Guger et al (1999); Dornhege et al (2004a); Lal et al (2004).…”

“…First, the regularization parameter is optimized using 10-fold cross validation within the training trial subset. Then we employ the technique of recursive channel elimination (RCE) first described by Lal et al (2004). This is a variant of recursive feature elimination (RFE), an embedded feature-selection method proposed by Guyon et al (2002) which an SVM is trained, its resulting weight vector is examined, and the subset of features with the lowest sum squared weight is eliminated (the features being grouped, in our case, into subsets corresponding to channels).…”