Find more posts tagged with
Sort by:
1 - 2 of
21
Often when approaching an unsupervised problem, it is easiest to use simple filter-based feature selection (like Remove Correlated Attributes) or variance-based methods (like PCA). The lack of a defined label or performance to optimize will make it more difficult to use the wrapper methods (like Evolutionary FS or Forward Selection or Backward Elimination), although there are some tricks to using those methods. Still, if I had a dataset with many attributes I would probably try one of the first two methods I listed first.
feature selection depends on the kind of data you have. If you are not sure about it you can try many combinations.
Although, when you select a model, you can see on your window "Help" section where there is explanation about all features of every model.
Hope it helps.