I've got 3070 training examples, including 1801 negative and 1269 positive. Your suggestions are very welcome.

Thanks
Alireza

Steven Barrett wrote:

 
Alireza,

....just one question - are you using a balanced training set ?

Steven
 
 
 

"Martin Sewell" <[log in to unmask]> Sent by: "The Support Vector Machine discussion list" <[log in to unmask]> 23-Apr-2002 20:46 Please respond to "The Support Vector Machine discussion list" <[log in to unmask]>

To:        SUPPORT-VECTOR-MACHINES         cc:          Subject:        Re: Help

Alireza Osareh asked me to forward the following message to this list.

At 09:54 23/04/2002 +0100, Alireza Osareh wrote:
>Hello,
>
>I have a classification problem and used Kernel-Adatron Algorithm
>without bias for the training purpose. I obtained reasonable results
>based on Gaussian kernel and also when I considered the same upper bound
>
>(C) restriction value to both (+1 (positive)) and (-1 (negative))
>classes.
>
>However, to control the sensitivity and specificity of the SVM,  I apply
>
>an upper bound only on (+1) class (C+) without any upper bound on the
>(-1) class, so I expect to have more specificity and less sensitivity
>compare to the initial results (and vice versa when considering
>restriction only on (-1) class).
>
>But, I do not know why the SVM give me always 100% Sensitivity and 0%
>Specificity, irrespective of the value of C+. To evaluate these results
>I assumed the output threshold equal to 0. In this case although I have
>examples from +1 and  -1 classes in my test set, the output is always a
>positive value for each test examples and therefore based on the output
>threshold  (0) I'll get 100% Sensitivity ....
>
>This case is also happened in the opposite way, when I put an upper
>bound on (-1) class and no restriction on (+1) class.
>
>Does anyone know why I've got these wrong results? Any idea will be
>greatly appreciated.
>
>Thanks a lot in advance.
>
>Alireza Assareh