Hello to all,
I've tried to use the K* weka alghorithm to solve a numeric prediction problem,
and it is seemed to me very good, while it is one of the few algorithms tha can
making numeric prediction working with both numerical and nominal attributes, so
I don't have the need to discretize the class before.
The problem is that I've seen that it's not possiblke to build a sort of
confusion matrix, and it's obvious, while it?s a numerical and not a class
problem. But I've the need to have an error-measure for each istance. I've seen
that it's possible to visualize all istances, and for each one to see the
effective value of the class and that of the predicted class.
I have for you some questions:
1) How can I compute classifier performances in this case?
2) How I should set, in the options window K* options in order to have a more
3) Is weka's grafic representation of test instances "editable" in any
Thanks to all
Show replies by date
On Fri, 23 Apr 2004, Alessandro asked about measuring classifier
performance when predicting a numerical attribute. One way would be r^2,
like in linear regression. I haven't checked whether Weka already knows
how to do this, or if you would have to write a (tiny) function.
I don't know the answers to the other questions.
a>1) How can I compute classifier performances in this case?
a>2) How I should set, in the options window K* options in order to have a more
a>3) Is weka's grafic representation of test instances "editable" in any