> Rajarshi Guha wrote: > > On Fri, 05 Mar 2004 11:42:24 +0000, Graham Jones wrote: > > > > > >>In article <[EMAIL PROTECTED]>, Rajarshi > >>Guha <[EMAIL PROTECTED]> writes > >> > >>>Hi, > >>> I've come across kernel density estimation technques and I was wonderin > >>>what can they be used for? It appears to me that they just give a > >>>representation of the PDF for the given data. But can these technique be > >>>used for other purposes? > >> > >>KDEs are often used for classification. > > > > > > Thanks for the pointers. > > Is it possible to use KDEs for multivariate regression ? > > > > Thanks > >
KDE is a really good one among several non-parametric density estimation methods. You can estimate either univariate or multivariate distributions using KED but you need to be cautious when estimating high-dimensional (higher than 5) PDF. Good references on KDE are (1) "Kernel smoothing" by Wand and Jones (1996) and (2) "density estimation and data analysis" (? I'm not sure) by Silverman (1986). Once you have underlying pdf of random variables, you can use many sophisticated theories based on pdfs in various areas. In pattern recognition for example, you can use estimated pdfs in classification (e.g., Bayesian classifier), clustering, and so on. For multivariate regression, I think I've heard or read about it somewhere but Don't remember exactly what and where it was. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
