I am the author of the OneR package and more and more people approach me asking how to use this classification algorithm for regression problems.
At the moment two approaches exist:
- either the target variable is discretized (binned) and intervals get predicted or
- there is only a limited number of values and the target becomes a categorical variable (a factor in R lingo)
My question
Which general strategies can be used to adapt a classification algorithm to regression problems. A simple possibility I see is compressing the abovementioned intervals into their means.
NB: I am using OneR in this question as a concrete example, the answer could be more general. Also references are highly welcome.
