Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!godot.cc.duq.edu!news.duke.edu!news.mathworks.com!newsfeed.internetmci.com!howland.reston.ans.net!torn!watserv3.uwaterloo.ca!watserv1!besner
From: besner@watserv1.uwaterloo.ca (Ken Seergobin)
Subject: Re: Pathology using NN
Message-ID: <Dq58ED.MCp@watserv1.uwaterloo.ca>
Sender: news@watserv1.uwaterloo.ca
Nntp-Posting-Host: watserv1.uwaterloo.ca
Organization: University of Waterloo
References: <3174470A.41C67EA6@camis.stanford.edu> <4l8bae$2q1@newsbf02.news.aol.com>
Date: Sat, 20 Apr 1996 04:27:49 GMT
Lines: 34

In article <4l8bae$2q1@newsbf02.news.aol.com>,
CarsonAD <carsonad@aol.com> wrote:
>My problem was
>to model how career counselors looked at profile data from tests and
>biodata and then made career recommendations to clients.  In particular, I
>used 9 input variables (aptitude, personality, interests, sex) and the
>criterion was recommendation to consider a career in science.  I trained a
>neural network (PNN, genetic adaptive) and had training set hit rate of
>89%, cross-validation to production set of 81%; a corresponding
>discriminant function analysis had a training set hit rate of 71%,
>cross-val. to 67%.  So the neural network appears to have done much


Hello,

What was the hit rate for the counselors?

Forgive me if I'm wrong, but is this type of modelling going
to be useful for young people looking for a carreer?  That is,
from my perspective we are not tailored to fit into *one*
carreer option.  In this instance isn't it enough that
someone has assessed the 'clients' aptitude and interest? 
- Both of which will act as a pointer.  Given the above
how does one determine a correct classification?  [With
regard to something like medical diagnosis the experimenter/
researcher more often than not should be able to - with
certainty - correct the the models classification in a 
principled manner during the training and building phase.]

Perhaps I should read your paper. 


Thanks,
Ken
