I'm experimenting with using CART (or maybe Random Forest) to classify genomic data. There are essentially two classes, whereof one is the 'normal' state and the other is the 'exceptional' state.
Now, I don't know how to properly set up a training set, because the 'normal' class appears with a much higher frequency (about factor 1000) than the 'exceptional' class.
In order to get the CART to work properly, do I rather use a 1:1 ratio of the samples in my training set or keep the 1000:1 ratio?,
I'm quite a beginner in this field, so any help is appreciated.