http://www.mathworks.com/matlabcentral/newsreader/view_thread/327869
MATLAB Central Newsreader  Performance function with pattern recognition in neural networks
Feed for thread: Performance function with pattern recognition in neural networks
enus
©19942014 by MathWorks, Inc.
webmaster@mathworks.com
MATLAB Central Newsreader
http://blogs.law.harvard.edu/tech/rss
60
MathWorks
http://www.mathworks.com/images/membrane_icon.gif

Wed, 27 Mar 2013 19:17:22 +0000
Performance function with pattern recognition in neural networks
http://www.mathworks.com/matlabcentral/newsreader/view_thread/327869#900988
William
Hi,<br>
<br>
A question for anyone who might know or have an opinion.<br>
<br>
I have set up a neural network to perform a pattern recognition (or classification) but I have found I am getting way too many false negatives compared to what I might actually get with say a Support Vector Machine set up. One possibility I am thinking is that the SVM set up can have harsh penalties for incorrect classifications. So, with this in mind, is there a "best" performance function for pattern recognition with neural networks?? Or am I best to say use use some function on the distance from the hyperplane (or similar)??<br>
<br>
Cheers

Fri, 29 Mar 2013 22:09:07 +0000
Re: Performance function with pattern recognition in neural networks
http://www.mathworks.com/matlabcentral/newsreader/view_thread/327869#901169
Greg Heath
"William " <william.henson@amec.com> wrote in message <kivgk2$i7$1@newscl01ah.mathworks.com>...<br>
> Hi,<br>
> <br>
> A question for anyone who might know or have an opinion.<br>
> <br>
> I have set up a neural network to perform a pattern recognition (or classification) but I have found I am getting way too many false negatives compared to what I might actually get with say a Support Vector Machine set up. One possibility I am thinking is that the SVM set up can have harsh penalties for incorrect classifications. So, with this in mind, is there a "best" performance function for pattern recognition with neural networks?? Or am I best to say use use some function on the distance from the hyperplane (or similar)??<br>
> <br>
> Cheers<br>
<br>
You have given absolutely no information that will let any one help you.<br>
Are you using patternnet with tansig/logsig or tansig/softmax ?<br>
Dimension of inputs? How many classes? For c classes does your target <br>
contain columns of the cdimensional unit matrix eye(c) or eye(c1)?<br>
How unbalanced is the data set: How large is each class? Are the ratios <br>
of class sizes the same as the apriori probabilities of the general population?<br>
Are the misclassification costs specified or the usual default values {0,1}?<br>
<br>
My apriori advice is to standardize your inputs and remove or modify <br>
outliers. Then use duplicates with or without added noise so that the <br>
number in each class is equal. If you have c classes, the cdimensional <br>
targets and class indices can be obtained from each other via ind2vec <br>
and vec2ind.<br>
<br>
Once the net is trained to yield approximately equal errors, you can <br>
transform the outputs by multiplying to account for differences in <br>
class priors and classification costs.<br>
<br>
You might find some old posts of mine in comp.ai.neuralnets and <br>
CSSM regarding priors and classification costs that will help.<br>
<br>
Hope tis helps.<br>
<br>
Greg