I want to train a binary classifier but I have an expected unbalanced dataset (90%/10%). Should I train an unbalanced dataset or to downsample the first class? This ratio is natural and will be always present in my data for prediction.
$\begingroup$
$\endgroup$
1
-
1$\begingroup$ See suggestions here: stats.stackexchange.com/questions/220026/… $\endgroup$– DaLCommented Sep 20, 2016 at 6:17
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Try and see which works better.
Given that you have a training set, a validation set and a metric that you want to optimize, you can answer all sorts of questions like this by simply trying.
