Dear Akbar i prefer you to use PCA principal Component Analysis.

Dear Saeedreza

thanks for your reply. But, as i mentioned i'm looking for a standard dataset, not dimension reduction method.

There are quite a few datasets that can give you a good visual representation of how well your reduction worked. The "swiss roll" is one example. Is this what you were looking for?

http://people.cs.uchicago.edu/~dinoj/manifold/swissroll.html

Dear Akbar Esmaeelzadeh,

Look the link. May be useful.

chrome-extension://oemmndcbldboiebfnladdacbdfmadadm/http://www.ijcaonline.org/volume13/number7/pxc3872471.pdf

chrome-extension://oemmndcbldboiebfnladdacbdfmadadm/http://webstaff.itn.liu.se/~jimjo/papers/infovis2009/johansso.pdf

chrome-extension://oemmndcbldboiebfnladdacbdfmadadm/http://arxiv.org/ftp/arxiv/papers/1110/1110.3569.pdf

Regards, Shafagat

I'm not aware of a standard, but it's easy to generate a k-dimensional sphere or ball or a mixture of Gaussians. I have used the values of features computed on a real image before. In one instance, PCA reduced a geometric feature set from about 1000 features down to 18 but the discriminating feature for my task was then found to be feature 15 of the 18. Basically, the data formed a solid sphere in most dimension combinations but was split in feature 15. There is good stuff in this research area.

Dear Dr. James.

Thank you for your answer. In my research interest, cardiac signal processing, there may be both k * n (k feature vector that each has 1 * n size that should be reduced down to 1 * L) and [ k * ( m * n)]  (k feature vector that each has m * n size that should be reduced to m * L) size datasets. I'm looking for such size datasets. Also, i mean the standard dataset, reference datasets that its results be referrable and publishable.

Regards,