FEATURESELECTIONBASEDON FISHERRATIOAND MUTUAL INFORMATION ANALYSESFOR ROBUST BRAINCOMPUTER INTERFACE

Cuntai Guan

Abstract

Thispaper proposes anovel feature selection method based ontwostage analysis ofFisher RatioandMutual Information forrobust Brain Computer Interface. Thismethoddecomposes multichannel brain signals into subbands. Thespatial filtering andfeature extraction isthenprocessed ineachsubband. Thetwo-stage analysis of Fisher Ratio andMutual Information iscarried outinthefeature domaintoreject thenoisy feature indexes andselect themostinformative combination fromtheremaining. Intheapproach, wedevelop twopractical solutions, avoiding thedifficulties ofusing highdimensional Mutual Information intheapplication, that arethefeature indexes clustering using cross Mutual Information andthelatter estimation based onconditional empirical PDF.Wetest theproposed feature selection method ontwoBCIdatasetsandtheresults are atleast comparable tothebestresults intheliterature. Themain advantage ofproposed method isthat themethod isfree fromany time-consuming parameter tweaking andtherefore suitable forthe BCIsystem design.
BibTeX
@inproceedings{guan2007featureselectionbasedon,
  author = {Cuntai Guan},
  title = {{FEATURESELECTIONBASEDON FISHERRATIOAND MUTUAL INFORMATION ANALYSESFOR ROBUST BRAINCOMPUTER INTERFACE}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2007},
  doi = {10.1109/icassp.2007.366685},
}

Loading related content...

Figure