in the motor imagery csp scenario provided we want to detect right or left hand imagiation using the algorithm. shouldn't it use a 3 class classification since there are 3 states : imagining right hand movement, imagining left hand movement and also imagining nothing.
wouldn't this make the scenario work inaccurately?
thanks
kiyarash
why a 2 class classification is used in motor imagery
-
- Posts: 775
- Joined: Tue Dec 04, 2012 3:53 pm
- Location: INRIA Rennes, FRANCE
Re: why a 2 class classification is used in motor imagery
Hi,
some research papers use an additional 'nothing' class, some do not. The Graz paradigm in openvibe is defined as two class; since during training and testing it is *assumed* that the user is a good boy/girl/person and obediently imagines of only left or right (as instructed) during the trial, this limited context is not requiring a third class. From real use perspective (for example controlling something), a third class of nothing could be useful. Adding a third class will also change the classification problem in a machine learning sense. How it will change it is doubtlessly a research question in itself. If anybody knows related links, feel free to post.
Cheers,
Jussi
some research papers use an additional 'nothing' class, some do not. The Graz paradigm in openvibe is defined as two class; since during training and testing it is *assumed* that the user is a good boy/girl/person and obediently imagines of only left or right (as instructed) during the trial, this limited context is not requiring a third class. From real use perspective (for example controlling something), a third class of nothing could be useful. Adding a third class will also change the classification problem in a machine learning sense. How it will change it is doubtlessly a research question in itself. If anybody knows related links, feel free to post.
Cheers,
Jussi