Thread Subject: Robust Mahalanobis Distance

Hi all,
  I just came across a paper discussing multivariate outlier detection and it talks about using minimum covariance determinant (MCD) estimator to calculate a robust mahalanobis distance then plotting against a chisquare . I found another post asking something similar and the response was that the wmulden function uses MCD but I have no idea how it can be implemented to calculated mahal distance. Anybody have any thoughts?
Thanks!!

>>help mahal

And you use mahal function. This is a dedicated matlab function that helps you to compute robust Mahalanobis distance

In order to find other distances type in command window:

>>help pdist

Or you implement the formulas from wikipedia in a matlab short script:

http://en.wikipedia.org/wiki/Mahalanobis_distance

Hope this all helps,

Michael

"Michelle " <gondu1@hotmail.com> wrote in message <h9rc8c$4r0$1@fred.mathworks.com>...
> Hi all,
> I just came across a paper discussing multivariate outlier detection and it talks about using minimum covariance determinant (MCD) estimator to calculate a robust mahalanobis distance then plotting against a chisquare . I found another post asking something similar and the response was that the wmulden function uses MCD but I have no idea how it can be implemented to calculated mahal distance. Anybody have any thoughts?
> Thanks!!

I have a similar problem like you 4 months ago.

Look at this thread:

http://www.mathworks.com/matlabcentral/newsreader/view_thread/250578#647624

Thanks! I was thrown by the "robust" in the mahalanobis description but now realized the mahal does what I need it to do
Thanks again!
Michelle


"Sprinceana " <mihaispr@yahoo.com> wrote in message <h9sube$h41$1@fred.mathworks.com>...
> "Michelle " <gondu1@hotmail.com> wrote in message <h9rc8c$4r0$1@fred.mathworks.com>...
> > Hi all,
> > I just came across a paper discussing multivariate outlier detection and it talks about using minimum covariance determinant (MCD) estimator to calculate a robust mahalanobis distance then plotting against a chisquare . I found another post asking something similar and the response was that the wmulden function uses MCD but I have no idea how it can be implemented to calculated mahal distance. Anybody have any thoughts?
> > Thanks!!
>
> I have a similar problem like you 4 months ago.
>
> Look at this thread:
>
> http://www.mathworks.com/matlabcentral/newsreader/view_thread/250578#647624