From: array chip <arrayprofile_at_yahoo.com>

Date: Thu, 07 Apr 2011 11:13:30 -0700 (PDT)

R-help_at_r-project.org mailing list

https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. Received on Thu 07 Apr 2011 - 18:15:19 GMT

Date: Thu, 07 Apr 2011 11:13:30 -0700 (PDT)

Dear all, this is not a pure R question, but really about how to set up a
multinomial logistic regression model to do a multi-class classification. I
would really appreciate if any of you would give me some of your thoughts and
recommendation.

Let's say we have 3-class classification problem: A, B and C. I have certain number of samples, with each sample, I have 3 variables (Xa, Xb and Xc). The trick here is that these 3 variables measures the extent of the likelihood of the samples being class A, B and C, i.e., Xa for class A, Xb for class B and Xc for class C. For a given sample i, we can simply make a rough prediction based on the values of Xa, Xb and Xc. For example:

for sample 1, Xa=10, Xb=50, Xc=15, then most likely I would predict sample 1 as class "B".

Then I have another set of variables Ya, Yb and Yc doing similar things.

I can construct a dataset as below:

Xa Xb Xc Ya Yb Yc class
sample 1 10 50 15 0.2 0.8 0.1 B
sample 2 8 4 6 0.7 0.5 0.3 A

*:
**:
*

and then make a model fit<-multinom(class~Xa+Xb+Xc+Ya+Yb+Yc)

But my understanding is that this model is not working in a way of by simply looking at each row of the data and pick the class that has the best Xs and/or Ys. In leave-one-out, sometimes it picks up a class that apparently is not a winner if I compare across Xs and Ys.

Greatly appreciate if anyone can suggest a more sensible way to construct the data and/or a different way of thinking of the problem at all.

John

[[alternative HTML version deleted]]

R-help_at_r-project.org mailing list

https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. Received on Thu 07 Apr 2011 - 18:15:19 GMT

Archive maintained by Robert King, hosted by
the discipline of
statistics at the
University of Newcastle,
Australia.

Archive generated by hypermail 2.2.0, at Thu 07 Apr 2011 - 18:20:28 GMT.

*
Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-help.
Please read the posting
guide before posting to the list.
*