I performed a wrapper random search feature selection with Weka as follows:
Evaluator: weka.attributeSelection.WrapperSubsetEval -B weka.classifiers.rules.ZeroR -F 5 -T 0.01 -R 1 --
Search:weka.attributeSelection.RandomSearch -F 25.0 -seed 1
my question is about how does it work, since Zero Rule returns just the mean or mode of the features, how can it be used as a predictor of the output variable.