Hi,
I'm looking for a java library for decision trees which accepts numeric attributes and classes/grades. Weka's J48 deals with discrete attributes but doesn't accept numeric ones.
Thanks
+2
A:
Not true, Weka's J48 (a variation of C4.5) does support numeric attributes. This is the main difference from the ID3 algorithm (among other things)
EDIT:
It appears you were referring to numeric output class rather than numeric input attributes. In this case, you are looking for model/regression trees. Weka has an implementation of the M5 algorithm by R. Quinlan (the same guy who originally introduced the C4.5 algorithm for discrete classes) in weka.classifiers.trees.M5P. The algorithm is similar to the usual C4.5 except it uses standard deviation instead of entropy, also the leaf nodes contain a regression model (or simply the mean).
Amro
2010-09-11 22:59:11
Right, but it doesn't support numeric classes, so you can't have a continuous spectrum of output from the model.
Ash
2010-09-12 21:53:33
@Ash: in that case you want model/regression trees, see the edit above...
Amro
2010-09-12 23:44:03