R help - Oct 2012 - Linear discriminant function analysis based median as group centroid and nonparametric scale estimators???

Dear All,

I am using a specific approach for my master thesis. In essence, a
supervised reclassification is used as an intermediate step to find chemical
parameters which are able to reclassify defined groups. These variables will
be used in a next step where location and scale estimators of the groups are
important. Traditionally linear discriminant analysis is used for
reclassification which uses group-means as centroid.

Caused by the characteristics of my data, I decided to use nonparametric
estimators, median as location and median absolute deviation as scale
estimator of the groups for the final step. For this purpose I would like to
know which chemical parameters are able to reclassify the defined groups
around the median instead of the mean. Therefore the ?traditional linear
discriminant analysis? seems to be not the best way for the supervised
reclassification.

Does somebody of you know if there is a package/function for R with a
?median based? discriminant analysis?
How would you handle this problem?
Does a robust lda or supervised reclassification exist which is suitable for
my problem?

I?m pleased about every suggestion.

Thank you and best regards
Thomas 






--
View this message in context:
http://r.789695.n4.nabble.com/Linear-discriminant-function-analysis-based-median-as-group-centroid-and-nonparametric-scale-estimat-tp4646915.html
Sent from the R help mailing list archive at Nabble.com.

Hi Thomas,

It does appear me and you are working on very similar sediment mixing model
problems. I ran a robust lda using the median by specifying method =
"t"
within the lda function.

So for my data I used:

model<-lda(All[,2:12],grouping=All$Catagory, method = "t")

Hope this helps.

Regards

Richard



--
View this message in context:
http://r.789695.n4.nabble.com/Linear-discriminant-function-analysis-based-median-as-group-centroid-and-nonparametric-scale-estimat-tp4646915p4647143.html
Sent from the R help mailing list archive at Nabble.com.