Prev: nonlinear regression by least square
Next: nonlinear regression by least square
From: Brian on 21 Jul 2010 11:24
Hi~

I am now doing a project about cancer detection.
I got data from 100 people and 600 parameters for each.
I also know that whether they are tumor or normal people.
I would like to select about 50 parameters which is the most significant in this detection.

I have written a program to kick out parameter one by one from 600 and the remaining 599 data are trained by using "neural network pattern recognition tool " (nprtool).

However, when I observe the 'confusion', the percentage varies a lot. For example, when I kick out the sixth parameter at the first time, the 'confusion' is about 30%. But when I kick out it again, the confusion is about 90%. It bothers me a lot and I don't know how to tackle it.

Besides, I have searched that support vector machine (svm) is feasible on my work.
It needs to classify the data in 1s or -1s. I don't understand how to determine whether the data is 1s or -1s.

I hope someone can answer my question!
Thanks!
 | 
Pages: 1
Prev: nonlinear regression by least square
Next: nonlinear regression by least square