Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!rochester!cornellcs!travelers.mail.cornell.edu!news.tc.cornell.edu!caen!hearst.acc.Virginia.EDU!murdoch!usenet
From: "Vijay S. Desai" <vsd3s@virginia.edu>
Subject: Number of output neurons in classification
X-Nntp-Posting-Host: bootp-92-200.bootp.virginia.edu
Message-ID: <DBq45z.Ht6@murdoch.acc.Virginia.EDU>
Sender: usenet@murdoch.acc.Virginia.EDU
Organization: University of Virginia
Date: Fri, 14 Jul 1995 20:41:58 GMT
Lines: 21

I am trying to use neural networks for a classification problem in
which I have two categories, say "good" and "bad". What are the pros
and cons in using a single neuron in the output layer compared to two
neurons? 

If I use only one neuron, I plan to use 0.5 as a cutoff level and then
if the output is less than 0.5 I will classify it as "bad" and "good" 
if it is greater than or equal to 0.5.

If I use two neurons, one for each category, than I will use the output
of the two neurons such that if either one of the neurons is greater
than or equal to 0.5, the particular observation belongs to that class.
and the observation is unclassifiable if both outputs are less than 0.5
 or both are greater than 0.5.

If now I want to extend the above scheme to three categories, say "good"
, "poor", and "bad", I can easily use three neurons, but I suppose it
will be difficult to work with one output neuron and three categories.

I will very much appreciate any comments any of you might have.

